content
stringlengths
73
1.12M
license
stringclasses
3 values
path
stringlengths
9
197
repo_name
stringlengths
7
106
chain_length
int64
1
144
<jupyter_start><jupyter_text>___ ___ Note- 1) First make the copy of the project in your drive then start with the solution. 2) Dont run the cell directly, first add a call above it then run the cell so that you dont miss the solution. # Logistic Regression Project In this project we will be working with a fake advertising data set, indicating whether or not a particular internet user clicked on an Advertisement. We will try to create a model that will predict whether or not they will click on an ad based off the features of that user. This data set contains the following features: * 'Daily Time Spent on Site': consumer time on site in minutes * 'Age': cutomer age in years * 'Area Income': Avg. Income of geographical area of consumer * 'Daily Internet Usage': Avg. minutes a day consumer is on the internet * 'Ad Topic Line': Headline of the advertisement * 'City': City of consumer * 'Male': Whether or not consumer was male * 'Country': Country of consumer * 'Timestamp': Time at which consumer clicked on Ad or closed window * 'Clicked on Ad': 0 or 1 indicated clicking on Ad ## Import Libraries **Import a few libraries you think you'll need (Or just import them as you go along!)**<jupyter_code>import pandas as pd import numpy as np import matplotlib.pyplot as plt import seaborn as sns<jupyter_output><empty_output><jupyter_text>## Get the Data **Read in the advertising.csv file and set it to a data frame called ad_data.**<jupyter_code>df = pd.read_csv('advertising.csv')<jupyter_output><empty_output><jupyter_text>**Check the head of ad_data**<jupyter_code>df.head() df.tail() df['Ad Topic Line'] ad = pd.get_dummies(df['Ad Topic Line']) ad<jupyter_output><empty_output><jupyter_text>** Use info and describe() on ad_data**<jupyter_code>df.info() df.describe() <jupyter_output><empty_output><jupyter_text>## Exploratory Data Analysis Let's use seaborn to explore the data! Try recreating the plots shown below! ** Create a histogram of the Age**<jupyter_code>sns.histplot(df['Age'] , bins= 30) <jupyter_output><empty_output><jupyter_text>**Create a jointplot showing Area Income versus Age.**<jupyter_code>sns.jointplot(x = 'Age' , y = 'Area Income' , data = df) <jupyter_output><empty_output><jupyter_text>** Create a jointplot of 'Daily Time Spent on Site' vs. 'Daily Internet Usage'**<jupyter_code>sns.jointplot(x = 'Daily Time Spent on Site' , y = 'Daily Internet Usage' , data = df , color = 'Green') <jupyter_output><empty_output><jupyter_text>** Finally, create a pairplot with the hue defined by the 'Clicked on Ad' column feature.**<jupyter_code>sns.pairplot(data = df ,palette = 'BuGn_r' , hue = 'Clicked on Ad') <jupyter_output><empty_output><jupyter_text># Logistic Regression Now it's time to do a train test split, and train our model! You'll have the freedom here to choose columns that you want to train on!** Split the data into training set and testing set using train_test_split**<jupyter_code>sns.heatmap(df.corr() , annot = True) from sklearn.model_selection import train_test_split df.info() x1 = df[['Age','Area Income','Daily Time Spent on Site','Daily Internet Usage','Male']] y1 = df['Clicked on Ad'] x1_train , x1_test , y1_train , y1_test = train_test_split(x1,y1,test_size = 0.3)<jupyter_output><empty_output><jupyter_text>** Train and fit a logistic regression model on the training set.**<jupyter_code>from sklearn.linear_model import LogisticRegression log_model = LogisticRegression() log_model.fit(x1_train , y1_train)<jupyter_output><empty_output><jupyter_text>## Predictions and Evaluations ** Now predict values for the testing data.**<jupyter_code>predict1 = log_model.predict(x1_test)<jupyter_output><empty_output><jupyter_text>** Create a classification report for the model.**<jupyter_code>from sklearn.metrics import confusion_matrix , classification_report confusion_matrix( y1_test , predict1) print(classification_report(y1_test, predict1)) <jupyter_output><empty_output>
no_license
/Logistic_Regression_Assignment.ipynb
young-ai-expert/Data-Science-Projects
12
<jupyter_start><jupyter_text># Vehicle detection and tracking project*picture by Udacity*<jupyter_code>import numpy as np import cv2 import glob import matplotlib.pyplot as plt import os import matplotlib.image as mpimg %matplotlib qt %matplotlib inline vehs = [] for image in os.listdir(os.getcwd() + "/vehicles/GTI_Right"): vehs.append(os.getcwd() + "/vehicles/GTI_Right/" + image) for image in os.listdir(os.getcwd() + "/vehicles/GTI_MiddleClose"): vehs.append(os.getcwd() + "/vehicles/GTI_MiddleClose/" + image) for image in os.listdir(os.getcwd() + "/vehicles/GTI_Left"): vehs.append(os.getcwd() + "/vehicles/GTI_Left/" + image) for image in os.listdir(os.getcwd() + "/vehicles/GTI_Far"): vehs.append(os.getcwd() + "/vehicles/GTI_Far/" + image) for image in os.listdir(os.getcwd() + "/vehicles/KITTI_extracted"): vehs.append(os.getcwd() + "/vehicles/KITTI_extracted/" + image) non_vehs = [] for image in os.listdir(os.getcwd() + "/non-vehicles/GTI"): non_vehs.append(os.getcwd() + "/non-vehicles/GTI/" + image) for image in os.listdir(os.getcwd() + "/non-vehicles/Extras"): non_vehs.append(os.getcwd() + "/non-vehicles/Extras/" + image) print(len(vehs)) print(len(non_vehs)) print(non_vehs[0]) testimg1 = mpimg.imread("test_images/test1.jpg") plt.imshow(testimg1) plt.show()<jupyter_output><empty_output><jupyter_text>## 1. Functions to extract the features### 1.1 get color features<jupyter_code>def color_hist(img, nbins=32, bins_range=(0, 256)): # Compute the histogram of the color channels separately channel1_hist = np.histogram(img[:, :, 0], bins=nbins, range=bins_range) channel2_hist = np.histogram(img[:, :, 1], bins=nbins, range=bins_range) channel3_hist = np.histogram(img[:, :, 2], bins=nbins, range=bins_range) # Concatenate the histograms into a single feature vector hist_features = np.concatenate((channel1_hist[0], channel2_hist[0], channel3_hist[0])) # Return the individual histograms, bin_centers and feature vector return hist_features def bin_spatial(img, size=(32, 32)): features = cv2.resize(img, size).ravel() # Return the feature vector return features<jupyter_output><empty_output><jupyter_text>### 1.2 get hog features<jupyter_code>from skimage.feature import hog def get_hog_features(img, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True): if vis == True: features, hog_image = hog(img, orientations=orient, pixels_per_cell=(pix_per_cell, pix_per_cell), cells_per_block=(cell_per_block, cell_per_block), transform_sqrt=True, visualise=vis, feature_vector=feature_vec) return features, hog_image else: features = hog(img, orientations=orient, pixels_per_cell=(pix_per_cell, pix_per_cell), cells_per_block=(cell_per_block, cell_per_block), transform_sqrt=True, visualise=vis, feature_vector=feature_vec) return features # example exp_img = cv2.cvtColor(mpimg.imread(vehs[0]), cv2.COLOR_RGB2GRAY) feats, hogImg = get_hog_features(exp_img, 8, 8, 2, vis=True, feature_vec=True) fig1 = plt.figure(figsize = (10,10)) ax1 = fig1.add_subplot(121) ax1.imshow(mpimg.imread(vehs[0]), interpolation='none') ax1.set_title('Initical picture') ax3 = fig1.add_subplot(122) ax3.imshow(hogImg, interpolation='none', cmap="gray") ax3.set_title('Image with HOG features') plt.show()<jupyter_output>C:\Anaconda3\envs\tensorflow-with-gpu-10\lib\site-packages\skimage\feature\_hog.py:119: skimage_deprecation: Default value of `block_norm`==`L1` is deprecated and will be changed to `L2-Hys` in v0.15 'be changed to `L2-Hys` in v0.15', skimage_deprecation) <jupyter_text>### 1.3 combine and normalize<jupyter_code>def extract_features(imgs, color_space='RGB', spatial_size=(32, 32), hist_bins=32, orient=9, pix_per_cell=8, cell_per_block=2, hog_channel=0, spatial_feat=True, hist_feat=True, hog_feat=True): # Create a list to append feature vectors to features = [] # Iterate through the list of images for file in imgs: file_features = [] # Read in each one by one image = mpimg.imread(file) # Apply color conversion if color_space != 'RGB': if color_space == 'HSV': feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2HSV) elif color_space == 'LUV': feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2LUV) elif color_space == 'HLS': feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2HLS) elif color_space == 'YUV': feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2YUV) elif color_space == 'YCrCb': feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2YCrCb) else: feature_image = np.copy(image) if spatial_feat == True: spatial_features = bin_spatial(feature_image, size=spatial_size) file_features.append(spatial_features) if hist_feat == True: hist_features = color_hist(feature_image, nbins=hist_bins) file_features.append(hist_features) if hog_feat == True: if hog_channel == 'ALL': hog_features = [] for channel in range(feature_image.shape[2]): hog_features.append(get_hog_features(feature_image[:, :, channel], orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True)) hog_features = np.ravel(hog_features) else: hog_features = get_hog_features(feature_image[:, :, hog_channel], orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True) # Append the new feature vector to the list of features file_features.append(hog_features) features.append(np.concatenate(file_features)) # Return list of feature vectors return features<jupyter_output><empty_output><jupyter_text>## 2. Search frames and find boxes**The following functions were built to evaluate and test the performance on single images, later on the functions are implemented in a class object with enhanced performance**### 2.1 sliding windows function<jupyter_code>def slide_window(img, x_start_stop=[None, None], y_start_stop=[None, None], xy_window=(64, 64), xy_overlap=(0.5, 0.5)): # If x and/or y start/stop positions not defined, set to image size if x_start_stop[0] == None: x_start_stop[0] = 0 if x_start_stop[1] == None: x_start_stop[1] = img.shape[1] if y_start_stop[0] == None: y_start_stop[0] = 0 if y_start_stop[1] == None: y_start_stop[1] = img.shape[0] # Region to be searched xspan = x_start_stop[1] - x_start_stop[0] yspan = y_start_stop[1] - y_start_stop[0] # Number of pixels per step in x/y nx_pix_per_step = np.int(xy_window[0] * (1 - xy_overlap[0])) ny_pix_per_step = np.int(xy_window[1] * (1 - xy_overlap[1])) # Number of windows in x/y nx_buffer = np.int(xy_window[0] * (xy_overlap[0])) ny_buffer = np.int(xy_window[1] * (xy_overlap[1])) nx_windows = np.int((xspan - nx_buffer) / nx_pix_per_step) ny_windows = np.int((yspan - ny_buffer) / ny_pix_per_step) # Initialize a list to append window parameters to window_list = [] for ys in range(ny_windows): for xs in range(nx_windows): # Calculate window position startx = xs * nx_pix_per_step + x_start_stop[0] endx = startx + xy_window[0] starty = ys * ny_pix_per_step + y_start_stop[0] endy = starty + xy_window[1] # Append window position to list window_list.append(((startx, starty), (endx, endy))) # Return the list of windows return window_list<jupyter_output><empty_output><jupyter_text>### 2.2 draw boxes function<jupyter_code>def draw_boxes(img, bboxes, color=(0, 0, 255), thick=6): # Copy the image imcopy = np.copy(img) # Iterate through the boxes for bbox in bboxes: # Draw a rectangle given bbox coordinates cv2.rectangle(imcopy, bbox[0], bbox[1], color, thick) # Return the image copy with boxes drawn return imcopy<jupyter_output><empty_output><jupyter_text>### 2.3 single image features extraction<jupyter_code>def single_img_features(img, color_space='RGB', spatial_size=(32, 32), hist_bins=32, orient=9, pix_per_cell=8, cell_per_block=2, hog_channel=0, spatial_feat=True, hist_feat=True, hog_feat=True): # Empty list img_features = [] # Apply color conversion if color_space != 'RGB': if color_space == 'HSV': feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2HSV) elif color_space == 'LUV': feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2LUV) elif color_space == 'HLS': feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2HLS) elif color_space == 'YUV': feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2YUV) elif color_space == 'YCrCb': feature_image = cv2.cvtColor(img, cv2.COLOR_RGB2YCrCb) else: feature_image = np.copy(img) # Compute spatial features if spatial_feat == True: spatial_features = bin_spatial(feature_image, size=spatial_size) # Append features to list img_features.append(spatial_features) # Compute histogram features if hist_feat == True: hist_features = color_hist(feature_image, nbins=hist_bins) # Append features to list img_features.append(hist_features) # Compute HOG features if hog_feat == True: if hog_channel == 'ALL': hog_features = [] for channel in range(feature_image.shape[2]): hog_features.extend(get_hog_features(feature_image[:,:,channel], orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True)) else: hog_features = get_hog_features(feature_image[:,:,hog_channel], orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True) # Append features to list img_features.append(hog_features) # Return concatenated array of features return np.concatenate(img_features)<jupyter_output><empty_output><jupyter_text>### Search windows function<jupyter_code> def search_windows(img, windows, clf, scaler, color_space='RGB', spatial_size=(32, 32), hist_bins=32, hist_range=(0, 256), orient=9, pix_per_cell=8, cell_per_block=2, hog_channel=0): # Create an empty list for the found windows on_windows = [] # Iterate over all windows in the windows list for window in windows: # Extract the test window from original image test_img = cv2.resize(img[window[0][1]:window[1][1], window[0][0]:window[1][0]], (64, 64)) # Extract features for that window using single_img_features() features = single_img_features(test_img, color_space=color_space, spatial_size=spatial_size, hist_bins=hist_bins, orient=orient, pix_per_cell=pix_per_cell, cell_per_block=cell_per_block, hog_channel=hog_channel, spatial_feat=spatial_feat, hist_feat=hist_feat, hog_feat=hog_feat) test_features = scaler.transform(np.array(features).reshape(1, -1)) # Predict using your classifier prediction = clf.predict(test_features) #If positive (prediction == 1) then save the window if prediction == 1: on_windows.append(window) # Return windows for positive detections return on_windows # RGB, HSV, LUV, HLS, YUV, YCrCb color_space = 'YCrCb' # HOG orientations orient = 9 # Pix per cell pix_per_cell = 8 # Cells per block cell_per_block = 2 # Hog channel (1,2,3 or ALL) hog_channel = "ALL" # Dimension for spatial binning spatial_size = (32, 32) # Number of histogram bins hist_bins = 32 spatial_feat = True hist_feat = True hog_feat = True y_start_stop = [350, None] hist_range = (0, 256)<jupyter_output><empty_output><jupyter_text>## 3. Training the classifier<jupyter_code># Get the features of cars and noncars car_features = extract_features(vehs, color_space, spatial_size, hist_bins, orient, pix_per_cell, cell_per_block, hog_channel) noncar_features = extract_features(non_vehs, color_space, spatial_size, hist_bins, orient, pix_per_cell, cell_per_block, hog_channel) # Check length of extracted image paths print(len(car_features)) print(len(noncar_features)) print(np.shape(car_features)) print(np.shape(noncar_features)) y = np.hstack((np.ones(len(car_features)), np.zeros(len(noncar_features)))) print(y.shape) X = np.vstack((car_features, noncar_features)).astype(np.float64) print(X.shape)<jupyter_output>(17760, 8460) <jupyter_text>#### Data normalization and classifier training<jupyter_code># Normalize training data from sklearn.preprocessing import StandardScaler X_Scaler = StandardScaler().fit(X) X_scaled = X_Scaler.transform(X) from sklearn.model_selection import train_test_split rand_state = np.random.randint(0, 100) # Split data in train/test data and shuffle it X_train, X_test, y_train, y_test = train_test_split(X_scaled, y, test_size = 0.2, random_state = rand_state) from sklearn.svm import LinearSVC # Train support vector machine svc = LinearSVC() svc.fit(X_train, y_train)<jupyter_output><empty_output><jupyter_text>### Get the accuarcy of the classifier<jupyter_code># Check the test accuarcy of the linear support vector machine svc.score(X_test, y_test)<jupyter_output><empty_output><jupyter_text>## 4. Implementation of sliding windows search (exemplary)<jupyter_code>image = mpimg.imread('test_images/test6.jpg') draw_image = np.copy(image) # Scale the image since its a .jpg image = image.astype(np.float32)/255 # Search with three different window sizes windows = slide_window(image, x_start_stop=[None, None], y_start_stop=y_start_stop, xy_window=(100, 100), xy_overlap=(0.5, 0.5)) windows2 = slide_window(image, x_start_stop=[None, None], y_start_stop=y_start_stop, xy_window=(200, 200), xy_overlap=(0.3, 0.3)) windows3 = slide_window(image, x_start_stop=[None, None], y_start_stop=y_start_stop, xy_window=(64, 64), xy_overlap=(0.3, 0.3)) # Get the found windows that match the features as list hot_windows = search_windows(image, windows, svc, X_Scaler, color_space=color_space, spatial_size=spatial_size, hist_bins=hist_bins, orient=orient, pix_per_cell=pix_per_cell, cell_per_block=cell_per_block, hog_channel=hog_channel) hot_windows2 = search_windows(image, windows2, svc, X_Scaler, color_space=color_space, spatial_size=spatial_size, hist_bins=hist_bins, orient=orient, pix_per_cell=pix_per_cell, cell_per_block=cell_per_block, hog_channel=hog_channel) hot_windows3 = search_windows(image, windows3, svc, X_Scaler, color_space=color_space, spatial_size=spatial_size, hist_bins=hist_bins, orient=orient, pix_per_cell=pix_per_cell, cell_per_block=cell_per_block, hog_channel=hog_channel) # Draw the found windows that match the features in boxes window_img = draw_boxes(draw_image, hot_windows, color=(0, 0, 255), thick=6) window_img2 = draw_boxes(window_img, hot_windows2, color=(0, 0, 255), thick=6) window_img3 = draw_boxes(window_img2, hot_windows2, color=(0, 0, 255), thick=6) plt.imshow(window_img3) # helper function for color conversion def convert_color(img, conv='RGB2YCrCb'): if conv == 'RGB2YCrCb': return cv2.cvtColor(img, cv2.COLOR_RGB2YCrCb) if conv == 'BGR2YCrCb': return cv2.cvtColor(img, cv2.COLOR_BGR2YCrCb) if conv == 'RGB2LUV': return cv2.cvtColor(img, cv2.COLOR_RGB2LUV) <jupyter_output><empty_output><jupyter_text>### 4.1 find cars function<jupyter_code># find cars function as shown in the lession # function uses a scale-factor to search with different window sizes # function as well replaces the overlapping with cells_per_steps def find_cars(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins): draw_img = np.copy(img) img = img.astype(np.float32)/255 img_tosearch = img[ystart:ystop,:,:] ctrans_tosearch = convert_color(img_tosearch, conv='RGB2YCrCb') boxes = [] if scale != 1: imshape = ctrans_tosearch.shape ctrans_tosearch = cv2.resize(ctrans_tosearch,(np.int(imshape[1]/scale),(np.int(imshape[0]/scale)))) ch1 = ctrans_tosearch[:,:,0] ch2 = ctrans_tosearch[:,:,1] ch3 = ctrans_tosearch[:,:,2] #Blocks and steps nxblocks = (ch1.shape[1] // pix_per_cell) - cell_per_block + 1 nyblocks = (ch1.shape[0] // pix_per_cell) - cell_per_block + 1 nfeat_per_block = orient * cell_per_block**2 window = 64 nblocks_per_window = (window // pix_per_cell) - cell_per_block + 1 # Replacing overlapping with cells_per_step cells_per_step = 2 nxsteps = (nxblocks - nblocks_per_window) // cells_per_step nysteps = (nyblocks - nblocks_per_window) // cells_per_step #get hog features hog1 = get_hog_features(ch1, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=False) hog2 = get_hog_features(ch2, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=False) hog3 = get_hog_features(ch3, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=False) for xb in range(nxsteps): for yb in range(nysteps): ypos = yb * cells_per_step xpos = xb * cells_per_step # Extract hog features hog_feat1 = hog1[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel() hog_feat2 = hog2[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel() hog_feat3 = hog3[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel() hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3)) x_left = xpos * pix_per_cell y_top = ypos * pix_per_cell # Extract img patch subimg = cv2.resize(ctrans_tosearch[y_top:y_top+window, x_left:x_left+window],(64,64)) spatial_features = bin_spatial(subimg, size=spatial_size) hist_features = color_hist(subimg, nbins=hist_bins) #test_features2 = np.concatenate((spatial_features, hist_features, hog_features)) #test_features = X_scaler.transform(np.array(test_features2)).reshape(1, -1) test_features = X_scaler.transform(np.hstack((spatial_features, hist_features, hog_features)).reshape(1, -1)) test_prediction = svc.predict(test_features) if test_prediction == 1: xbox_left = np.int(x_left * scale) ytop_draw = np.int(y_top * scale) win_draw = np.int(window*scale) cv2.rectangle(draw_img,(xbox_left, ytop_draw+ystart),(xbox_left+win_draw, ytop_draw+ win_draw+ystart),(0,0,255),6) boxes.append(((xbox_left, ytop_draw+ystart),(xbox_left+win_draw, ytop_draw+ win_draw+ystart))) return draw_img, boxes ystart = 400 ystop = 656 scale = 1.5 img_test = mpimg.imread('test_images/test1.jpg') out_img, boxes = find_cars(img_test, ystart, ystop, scale, svc, X_Scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins) plt.imshow(out_img) print(boxes)<jupyter_output>[((48, 424), (144, 520)), ((792, 400), (888, 496)), ((816, 400), (912, 496)), ((840, 400), (936, 496)), ((864, 400), (960, 496)), ((1056, 400), (1152, 496)), ((1128, 400), (1224, 496)), ((1128, 424), (1224, 520)), ((1152, 400), (1248, 496)), ((1152, 424), (1248, 520))] <jupyter_text>### 4.2 Add heatmap to cut out false-positives<jupyter_code># Heatmap functions / False positives threshold = 1 heat = np.zeros_like(img_test[:,:,0]).astype(np.float) def add_heat(heatmap, bbox_list): for box in bbox_list: heatmap[box[0][1]:box[1][1], box[0][0]:box[1][0]] += 1 return heatmap def apply_threshold(heatmap, threshold): heatmap[heatmap <= threshold] = 0 return heatmap def draw_labeled_boxes(img, labels): for car_number in range(1, labels[1]+1): nonzero = (labels[0] == car_number).nonzero() nonzeroy = np.array(nonzero[0]) nonzerox = np.array(nonzero[1]) bbox = ((np.min(nonzerox), np.min(nonzeroy)), (np.max(nonzerox), np.max(nonzeroy))) #print(bbox[0], bbox[1]) #cv2.rectangle(img, bbox[0], bbox[1], (0,0,255), 6) img = draw_boxes(img, [bbox], color=(0, 0, 255), thick=6) return img heat = add_heat(heat, boxes) heat= apply_threshold(heat, threshold) heatmap = np.clip(heat, 0, 255) from scipy.ndimage.measurements import label labels = label(heatmap) print(labels[1], 'vehicles found') plt.imshow(labels[0], cmap='hot') draw_img2 = draw_labeled_boxes(np.copy(img_test), labels) plt.imshow(draw_img2) # pulling everything together threshold = 1 def vehicle_detection(image, ystart, ystop, scale, svc, X_Scaler, orient,pix_per_cell, cell_per_block, spatial_size, hist_bins, threshold): #find cars in image out_img, boxes = find_cars(image, ystart, ystop, scale, svc, X_Scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins) heat = np.zeros_like(img_test[:,:,0]).astype(np.float) box_list = boxes heat = add_heat(heat, box_list) heat= apply_threshold(heat, threshold) heatmap = np.clip(heat, 0, 255) labels = label(heatmap) draw_img = draw_labeled_boxes(np.copy(image), labels) return draw_img draw_img = vehicle_detection(np.copy(img_test), ystart, ystop, scale, svc, X_Scaler, orient,pix_per_cell, cell_per_block, spatial_size, hist_bins, threshold) plt.imshow(draw_img) from moviepy.editor import VideoFileClip from IPython.display import HTML def process_image(image): #image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) processed_img = vehicle_detection(image, ystart, ystop, scale, svc, X_Scaler, orient,pix_per_cell, cell_per_block, spatial_size, hist_bins, threshold) return processed_img #smooth the pipeline outcome def find_cars_multiscale(img, ystart_ystop_scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins): draw_img = np.copy(img) img = img.astype(np.float32)/255 for (ystart, ystop, scale) in ystart_ystop_scale: img_tosearch = img[ystart:ystop,:,:] ctrans_tosearch = convert_color(img_tosearch, conv='RGB2YCrCb') boxes = [] if scale != 1: imshape = ctrans_tosearch.shape ctrans_tosearch = cv2.resize(ctrans_tosearch,(np.int(imshape[1]/scale),(np.int(imshape[0]/scale)))) ch1 = ctrans_tosearch[:,:,0] ch2 = ctrans_tosearch[:,:,1] ch3 = ctrans_tosearch[:,:,2] # Blocks and steps nxblocks = (ch1.shape[1] // pix_per_cell) - cell_per_block + 1 nyblocks = (ch1.shape[0] // pix_per_cell) - cell_per_block + 1 nfeat_per_block = orient * cell_per_block**2 window = 64 nblocks_per_window = (window // pix_per_cell) - cell_per_block + 1 cells_per_step = 2 nxsteps = (nxblocks - nblocks_per_window) // cells_per_step nysteps = (nyblocks - nblocks_per_window) // cells_per_step # Extract hog features hog1 = get_hog_features(ch1, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=False) hog2 = get_hog_features(ch2, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=False) hog3 = get_hog_features(ch3, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=False) for xb in range(nxsteps): for yb in range(nysteps): ypos = yb * cells_per_step xpos = xb * cells_per_step # Extract hog features hog_feat1 = hog1[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel() hog_feat2 = hog2[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel() hog_feat3 = hog3[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel() hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3)) x_left = xpos * pix_per_cell y_top = ypos * pix_per_cell #extract img patch subimg = cv2.resize(ctrans_tosearch[y_top:y_top+window, x_left:x_left+window],(64,64)) spatial_features = bin_spatial(subimg, size=spatial_size) hist_features = color_hist(subimg, nbins=hist_bins) #test_features2 = np.concatenate((spatial_features, hist_features, hog_features)) #test_features = X_scaler.transform(np.array(test_features2)).reshape(1, -1) test_features = X_scaler.transform(np.hstack((spatial_features, hist_features, hog_features)).reshape(1, -1)) test_prediction = svc.predict(test_features) if test_prediction == 1: xbox_left = np.int(x_left * scale) ytop_draw = np.int(y_top * scale) win_draw = np.int(window*scale) cv2.rectangle(draw_img,(xbox_left, ytop_draw+ystart),(xbox_left+win_draw, ytop_draw+ win_draw+ystart),(0,0,255),6) boxes.append(((xbox_left, ytop_draw+ystart),(xbox_left+win_draw, ytop_draw+ win_draw+ystart))) return draw_img, boxes from collections import deque heatmaps = deque(maxlen=3) threshold = 2 ystart_ystop_scale = [(400, 560, 1.5), (450, 650, 1.8), (500, 700, 2.5)] def vehicle_detection_smooth(image, ystart_ystop_scale, svc, X_Scaler, orient,pix_per_cell, cell_per_block, spatial_size, hist_bins, threshold): # find cars in image out_img, boxes = find_cars_multiscale(image, ystart_ystop_scale, svc, X_Scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins) heat = np.zeros_like(img_test[:,:,0]).astype(np.float) box_list = boxes current_heatmap = add_heat(image, box_list) heatmaps.append(current_heatmap) heatmap_sum = sum(heatmaps) heat = apply_threshold(heatmap_sum, threshold) heatmap = np.clip(heat, 0, 255) labels = label(heatmap) draw_img = draw_labeled_boxes(np.copy(image), labels) return draw_img draw_img = vehicle_detection_smooth(np.copy(img_test), ystart_ystop_scale, svc, X_Scaler, orient,pix_per_cell, cell_per_block, spatial_size, hist_bins, threshold) plt.imshow(draw_img)<jupyter_output><empty_output><jupyter_text>## 5. VehicleDetector Class Object**This class object combines the prework described above in one class object**<jupyter_code># To make thing easier the following class object, comprising the relevant functions to detect vehicles, is created. class VehicleDetector: def __init__(self): # Init the class using the parameters as for the single functions before self.color_space = 'YCrCb' self.orient = 9 self.pix_per_cell = 8 self.cell_per_block = 2 self.hog_channel = "ALL" self.spatial_size = (32, 32) self.hist_bins = 32 self.spatial_feat = True self.hist_feat = True self.hog_feat = True # Current heatmap self.heatmap = None # Heatmaps of last 3 frames self.heat_images = deque(maxlen=3) self.frame_count = 0 self.full_frame_processing_interval = 10 # Xstart self.xstart = 600 # Various Scales self.ystart_ystop_scale = None # Kernal For Dilation self.kernel = np.ones((50, 50)) # Threshold for Heatmap self.threshold = None self.X_scaler = X_Scaler self.svc = svc def find_cars_smooth(self, img): X_scaler = self.X_scaler orient = self.orient pix_per_cell = self.pix_per_cell cell_per_block = self.cell_per_block spatial_size = self.spatial_size hist_bins = self.hist_bins svc = self.svc box_list = [] draw_img = np.copy(img) img = img.astype(np.float32) / 255 if self.frame_count % self.full_frame_processing_interval == 0: mask = np.ones_like(img[:, :, 0]) else: mask = np.sum(np.array(self.heat_images), axis=0) mask[(mask > 0)] = 1 mask = cv2.dilate(mask, self.kernel, iterations=1) self.frame_count += 1 for (ystart, ystop, scale) in self.ystart_ystop_scale: nonzero = mask.nonzero() nonzeroy = np.array(nonzero[0]) nonzerox = np.array(nonzero[1]) if len(nonzeroy) != 0: ystart = max(np.min(nonzeroy), ystart) ystop = min(np.max(nonzeroy), ystop) if len(nonzeroy) != 0: xstart = max(np.min(nonzerox), self.xstart) xstop = np.max(nonzerox) else: continue if xstop <= xstart or ystop <= ystart: continue img_tosearch = img[ystart:ystop, xstart:xstop, :] ctrans_tosearch = convert_color(img_tosearch, conv='RGB2YCrCb') if scale != 1: imshape = ctrans_tosearch.shape ys = np.int(imshape[1] / scale) xs = np.int(imshape[0] / scale) if (ys < 1 or xs < 1): continue ctrans_tosearch = cv2.resize(ctrans_tosearch, (np.int(imshape[1] / scale), np.int(imshape[0] / scale))) if ctrans_tosearch.shape[0] < 64 or ctrans_tosearch.shape[1] < 64: continue ch1 = ctrans_tosearch[:, :, 0] ch2 = ctrans_tosearch[:, :, 1] ch3 = ctrans_tosearch[:, :, 2] nxblocks = (ch1.shape[1] // pix_per_cell) - 1 nyblocks = (ch1.shape[0] // pix_per_cell) - 1 nfeat_per_block = orient * cell_per_block ** 2 window = 64 nblocks_per_window = (window // pix_per_cell) - 1 cells_per_step = 2 nxsteps = (nxblocks - nblocks_per_window) // cells_per_step nysteps = (nyblocks - nblocks_per_window) // cells_per_step # Compute individual channel HOG features for the entire image hog1 = get_hog_features(ch1, orient, pix_per_cell, cell_per_block, feature_vec=False) hog2 = get_hog_features(ch2, orient, pix_per_cell, cell_per_block, feature_vec=False) hog3 = get_hog_features(ch3, orient, pix_per_cell, cell_per_block, feature_vec=False) for xb in range(nxsteps + 1): for yb in range(nysteps + 1): ypos = yb * cells_per_step xpos = xb * cells_per_step # Extract HOG for this patch hog_feat1 = hog1[ypos:ypos + nblocks_per_window, xpos:xpos + nblocks_per_window].ravel() hog_feat2 = hog2[ypos:ypos + nblocks_per_window, xpos:xpos + nblocks_per_window].ravel() hog_feat3 = hog3[ypos:ypos + nblocks_per_window, xpos:xpos + nblocks_per_window].ravel() hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3)) xleft = xpos * pix_per_cell ytop = ypos * pix_per_cell # Extract the image patch subimg = ctrans_tosearch[ytop:ytop + window, xleft:xleft + window] # Get color features spatial_features = bin_spatial(subimg, size=spatial_size) hist_features = color_hist(subimg, nbins=hist_bins) # Scale features and make a prediction test_features = X_scaler.transform( np.hstack((spatial_features, hist_features, hog_features)).reshape(1, -1)) # test_features = X_scaler.transform(np.hstack((shape_feat, hist_feat)).reshape(1, -1)) test_prediction = svc.predict(test_features) if test_prediction == 1: xbox_left = xstart + np.int(xleft * scale) ytop_draw = np.int(ytop * scale) win_draw = np.int(window * scale) box_list.append( ((xbox_left, ytop_draw + ystart), (xbox_left + win_draw, ytop_draw + win_draw + ystart))) # Add heat to each box in box list self.add_heatmap_and_threshold(draw_img, box_list, self.threshold) # Find final boxes from heatmap using label function labels = label(self.heatmap) VehicleDetector.draw_labeled_bboxes_smooth(draw_img, labels) return draw_img def add_heatmap_and_threshold(self, draw_img, bbox_list, threshold): heatmap = np.zeros_like(draw_img[:, :, 0]).astype(np.float) for box in bbox_list: # Add += 1 for all pixels inside each bbox heatmap[box[0][1]:box[1][1], box[0][0]:box[1][0]] += 1 self.heat_images.append(heatmap) self.heatmap = np.sum(np.array(self.heat_images), axis=0) self.heatmap[self.heatmap <= threshold] = 0 @staticmethod def draw_labeled_bboxes_smooth(img, labels): # Iterate through all detected cars for car_number in range(1, labels[1] + 1): # Find pixels with each car_number label value nonzero = (labels[0] == car_number).nonzero() # Identify x and y values nonzeroy = np.array(nonzero[0]) nonzerox = np.array(nonzero[1]) # Box based on min/max x and y bbox = ((np.min(nonzerox), np.min(nonzeroy)), (np.max(nonzerox), np.max(nonzeroy))) # Draw the box on the image cv2.rectangle(img, bbox[0], bbox[1], (0, 0, 255), 6)<jupyter_output><empty_output><jupyter_text>## 6. Combining it with my last project submission<jupyter_code>import pickle with open('objs.pkl', 'rb') as f: mtx, dist = pickle.load(f) def undistort_img(img, mtx, dist): img_undist = cv2.undistort(img, mtx, dist, None, mtx) return img_undist def grayscale(img): img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) return img_gray def SobelOp(img, kernelSize, thresh_min, thresh_max): sobelx = cv2.Sobel(img, cv2.CV_64F, 1, 0, ksize = kernelSize) sobely = cv2.Sobel(img, cv2.CV_64F, 0, 1, ksize = kernelSize) abs_sobelx = np.absolute(sobelx) scaled_sobel = np.uint8(255*abs_sobelx/np.max(abs_sobelx)) sxbinary = np.zeros_like(scaled_sobel) sxbinary[(scaled_sobel > thresh_min) & (scaled_sobel <= thresh_max)] = 1 return scaled_sobel, sxbinary def HLSconv(img, s_thresh_min, s_thresh_max): # convert to hls hls = cv2.cvtColor(img, cv2.COLOR_RGB2HLS) s_chan = hls[:,:,2] s_binary = np.zeros_like(s_chan) s_binary[(s_chan > s_thresh_min) & (s_chan <= s_thresh_max)] = 1 return s_binary def StackFilters(sxbinary, s_binary): color_binary = np.dstack((np.zeros_like(sxbinary), sxbinary, s_binary))*255 comb_binary = np.zeros_like(sxbinary) comb_binary[(s_binary == 1) | (sxbinary == 1)] = 1 return comb_binary def IMGwarp(img): src = np.float32([[740, 460], [1050, 680], [280, 680], [580, 460]]) dst = np.float32([[950, 0], [950, 720], [370, 720], [370, 0]]) img_size = (img.shape[1], img.shape[0]) M = cv2.getPerspectiveTransform(src, dst) Minv = cv2.getPerspectiveTransform(dst, src) img_warped = cv2.warpPerspective(img, M, img_size, flags=cv2.INTER_LINEAR) return M, Minv, img_warped def maskImage(img_w): mask = np.zeros_like(img_w) ignore_mask_color = 255 ignore_mask_color2 = 0 vertices = np.array([[(200, 720),(300, 200), (1100, 200), (1250,720)]], dtype=np.int32) vertices2 = np.array([[(600, 720),(675, 450), (725, 450), (800,720)]], dtype=np.int32) cv2.fillPoly(mask, vertices, ignore_mask_color) cv2.fillPoly(mask, vertices2, ignore_mask_color2) img_masked = cv2.bitwise_and(img_w, mask) return img_masked def FindLanes(img_warped): histogram = np.sum(img_warped[img_warped.shape[0]//2:,:], axis=0) out_img = np.dstack((img_warped, img_warped, img_warped))*255 midpoint = np.int(histogram.shape[0]/2) leftx_base = np.argmax(histogram[:midpoint]) rightx_base = np.argmax(histogram[midpoint:])+midpoint offset_shadow = 100 detected_left = False detected_right = False if (leftx_base > 300) & (leftx_base < 500): detected_left = True if (rightx_base > 800) & (rightx_base < 1000): detected_right = True if detected_left == False: leftx_base = np.argmax(histogram[:midpoint-offset_shadow]) if detected_right == False: rightx_base = np.argmax(histogram[midpoint+offset_shadow:])+midpoint return midpoint, leftx_base, rightx_base, detected_left, out_img def distanceMid(img2, right_fitx, left_fitx): camera_position = img2.shape[1]/2 lane_center = (right_fitx[719] + left_fitx[719])/2 center_offset_pixels = abs(camera_position - lane_center) center_offset = center_offset_pixels*xm_per_pix return center_offset def WindowSearch(img_warped, n_win, marg, minpix, leftx_base, rightx_base, out_img): nwindows = n_win #9 window_height = np.int(img_warped.shape[0]/nwindows) nonzero = img_warped.nonzero() nonzeroy = np.array(nonzero[0]) nonzerox = np.array(nonzero[1]) leftx_current = leftx_base rightx_current = rightx_base margin = marg minpix = minpix left_lane_inds = [] right_lane_inds = [] for window in range(nwindows): win_y_low = img_warped.shape[0] - (window + 1) * window_height win_y_high = img_warped.shape[0] - window * window_height win_x_left_low = leftx_current - margin win_x_left_high = leftx_current + margin win_x_right_low = rightx_current - margin win_x_right_high = rightx_current + margin cv2.rectangle(out_img, (win_x_left_low, win_y_low),(win_x_left_high, win_y_high), (0, 255, 0), 2) cv2.rectangle(out_img, (win_x_right_low, win_y_low),(win_x_right_high, win_y_high), (0, 255, 0), 2) good_left_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) & (nonzerox >= win_x_left_low) & (nonzerox < win_x_left_high)).nonzero()[0] good_right_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) & (nonzerox >= win_x_right_low) & (nonzerox < win_x_right_high)).nonzero()[0] left_lane_inds.append(good_left_inds) right_lane_inds.append(good_right_inds) if (len(good_left_inds) > minpix) & (len(good_right_inds) > minpix): leftx_current = np.int(np.mean(nonzerox[good_left_inds])) rightx_current = np.int(np.mean(nonzerox[good_right_inds])) # concatenate found lane indices left_lane_inds = np.concatenate(left_lane_inds) right_lane_inds = np.concatenate(right_lane_inds) # extract the pixel positions left_x = nonzerox[left_lane_inds] left_y = nonzeroy[left_lane_inds] right_x = nonzerox[right_lane_inds] right_y = nonzeroy[right_lane_inds] # fit second order polynomial left_fit = np.polyfit(left_y, left_x, 2) right_fit = np.polyfit(right_y, right_x, 2) # Visualize sliding windows ploty = np.linspace(0, img_warped.shape[0]-1, img_warped.shape[0]) left_fitx = left_fit[0]*ploty**2 + left_fit[1] * ploty + left_fit[2] right_fitx = right_fit[0]*ploty**2 + right_fit[1] * ploty + right_fit[2] return left_lane_inds, right_lane_inds, left_fitx, right_fitx, ploty, left_x, left_y, right_x, right_y def curvature(left_x, left_y, right_x, right_y, ploty): lane_length = 30 lane_width = 3.7 y_eval = np.max(ploty) ym_per_pix = lane_length/720 xm_per_pix = lane_width/580 left_fit_cr = np.polyfit(left_y * ym_per_pix, left_x * xm_per_pix, 2) right_fit_cr = np.polyfit(right_y * ym_per_pix, right_x * xm_per_pix, 2) left_curverad_cr = ((1 + (2 * left_fit_cr[0] * y_eval*ym_per_pix + left_fit_cr[1]) ** 2) ** 1.5) / np.absolute(2 * left_fit_cr[0]) right_curverad_cr = ((1 + (2 * right_fit_cr[0] * y_eval*ym_per_pix + right_fit_cr[1]) ** 2) ** 1.5) / np.absolute(2 * right_fit_cr[0]) return left_curverad_cr, right_curverad_cr def lanePosition(): left_pos = (midpoint - leftx_base) * xm_per_pix right_pos = (rightx_base - midpoint) * xm_per_pix line_base_pos = np.min([left_pos, right_pos]) return left_pos, right_pos, line_base_pos def warpBack(img_warped, left_fitx, right_fitx, ploty, Minv, image_undist): warp_zero = np.zeros_like(img_warped).astype(np.uint8) color_warp = np.dstack((warp_zero, warp_zero, warp_zero)) # Recast the points pts_left = np.array([np.transpose(np.vstack([left_fitx, ploty]))]) pts_right = np.array([np.flipud(np.transpose(np.vstack([right_fitx, ploty])))]) pts = np.hstack((pts_left, pts_right)) # Draw the lane cv2.fillPoly(color_warp, np.int_([pts]), (0, 255, 0)) # Warp back to original image newwarp = cv2.warpPerspective(color_warp, Minv, (image_undist.shape[1], image_undist.shape[0])) result = cv2.addWeighted(image_undist, 1, newwarp, 0.3, 0) return result def select_yellow(image_in): image_in2 = cv2.cvtColor(image_in, cv2.COLOR_BGR2RGB) hsv = cv2.cvtColor(image_in2, cv2.COLOR_RGB2HSV) lower = np.array([20,60,60]) upper = np.array([38,174, 250]) mask = cv2.inRange(hsv, lower, upper) return mask def select_white(image_in): lower = np.array([202,202,202]) upper = np.array([255,255,255]) mask = cv2.inRange(image_in, lower, upper) return mask def comb_thresh(image_in): yellow = select_yellow(image_in) white = select_white(image_in) combined_binary = np.zeros_like(yellow) combined_binary[(yellow >= 1) | (white >= 1)] = 1 return combined_binary def pipeline3(image, mtx, dist, kernelSize): out1 = undistort_img(image, mtx, dist) out2 = comb_thresh(out1) M, Minv, out3 = IMGwarp(out2) mid, leftxbase, rightxbase, detected, out_img = FindLanes(out3) left_inds, right_inds, leftfit, rightfit, plotyy, leftx, lefty, rightx, righty = WindowSearch(out3, 9, 100, 10, leftxbase, rightxbase, out_img) left_curverad_cr, right_curverad_cr = curvature(leftx, lefty, rightx, righty, plotyy) left_fitx = leftfit right_fitx = rightfit global l_fit_buffer global r_fit_buffer global old_img_lines if old_img_lines is None: old_img_lines = out2 # Compare new frame to previous frame ret = cv2.matchShapes(old_img_lines, out2, 1, 0.0) if ret < 50: old_img_lines = out2 if l_fit_buffer is None: l_fit_buffer = np.array([left_fitx]) if r_fit_buffer is None: r_fit_buffer = np.array([right_fitx]) l_fit_buffer = np.append(l_fit_buffer, [left_fitx], axis=0)[-filter_size:] r_fit_buffer = np.append(r_fit_buffer, [right_fitx], axis=0)[-filter_size:] # Compute the mean l_fit_mean = np.mean(l_fit_buffer, axis=0) r_fit_mean = np.mean(r_fit_buffer, axis=0) result2 = warpBack(out3, leftfit, rightfit, plotyy, Minv, out1) return result2 vehicleDetector = VehicleDetector() vehicleDetector.ystart_ystop_scale = [(380, 480, 1), (400, 600, 1.5), (500, 700, 2.5)] vehicleDetector.threshold = 2 image3 = mpimg.imread('test_images/test1.jpg') output3 = vehicleDetector.find_cars_smooth(image3) plt.figure() plt.imshow(output3) plt.title('Car Positions') filter_size = 10 old_img_lines = None l_fit_buffer = None r_fit_buffer = None def process_image_lanes(image): image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) processed_img = pipeline3(image, mtx, dist, 5) return processed_img #combine both vehicleDetector = VehicleDetector() vehicleDetector.ystart_ystop_scale = [(380, 480, 1), (400, 600, 1.5), (500, 700, 2.5)] vehicleDetector.threshold = 2 filter_size = 10 old_img_lines = None l_fit_buffer = None r_fit_buffer = None def process_image_comb(image): image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) processed_img = pipeline3(image, mtx, dist, 5) processed_img = cv2.cvtColor(processed_img, cv2.COLOR_RGB2BGR) processed_img2 = vehicleDetector.find_cars_smooth(processed_img) return processed_img2 project_video_output = './output_videos/project_video_output_comb.mp4' clip9000 = VideoFileClip("./project_video.mp4") white_clip9000 = clip9000.fl_image(process_image_comb) %time white_clip9000.write_videofile(project_video_output, audio=False)<jupyter_output>[MoviePy] >>>> Building video ./output_videos/project_video_output_comb.mp4 [MoviePy] Writing video ./output_videos/project_video_output_comb.mp4
no_license
/CarND-Vehicle-Detection-master-P5.ipynb
azumf/CarND-Vehicle-Detection-P5
16
<jupyter_start><jupyter_text># 교차검증 그리드 서치 (cross_val_score) p324~<jupyter_code>from sklearn.model_selection import cross_val_score import numpy as np best_score = 0 for gamma in [0.001, 0.01, 0.1, 1, 10, 100]: for C in [0.001, 0.01, 0.1, 1, 10, 100]: # 매개변수의 각 조합에 대해 SVC를 훈련시킵니다 svm = SVC(gamma=gamma, C=C) scores = cross_val_score(svm, X_trainval, y_trainval, cv=5) # print(scores) score=np.mean(scores) if score > best_score: # 점수가 더 높으면 매개변수와 함께 기록합니다 best_score = score best_parameters = {'C': C, 'gamma': gamma} # 훈련 세트와 검증 세트를 합쳐 모델을 다시 만든 후 테스트 세트를 사용해 평가 svm = SVC(**best_parameters) # **best_parameters : SVC(C=C, gamma=gamma) -> SVC(C=10, gamma=0.001) svm.fit(X_trainval, y_trainval) test_score = svm.score(X_test, y_test) print("검증 세트에서 최고 점수: {:.2f}".format(best_score)) print("최적 파라미터: ", best_parameters) print("최적 파라미터에서 테스트 세트 점수: {:.2f}".format(test_score)) from sklearn.model_selection import GridSearchCV param_grid = {'C':[0.001, 0.01, 0.1, 1, 10, 100],'gamma':[0.001, 0.01, 0.1, 1, 10, 100]} # 반복문 돌려 최적값 찾기 gsc = GridSearchCV(SVC(), param_grid, cv=5, return_train_score=True) gsc.fit(X_trainval, y_trainval) #(6개 * 6개 * 5번 반복) print(gsc.best_score_) print(gsc.best_params_) print(gsc.score(X_test, y_test)) import pandas as pd results = pd.DataFrame(gsc.cv_results_) # results.head() results.iloc[:, 4:].head() import mglearn # results['mean_test_score'] # 36개 값, scores = results['mean_test_score'].values.reshape(6,6) # print(scores) mglearn.tools.heatmap(scores, xlabel='gamma', ylabel='C', yticklabels=param_grid['C'], xticklabels=['gamma']) param_grid = [{'kernel': ['rbf'], 'C': [0.001, 0.01, 0.1, 1, 10, 100], 'gamma': [0.001, 0.01, 0.1, 1, 10, 100]}, {'kernel': ['linear'], 'C': [0.001, 0.01, 0.1, 1, 10, 100]}] grid_search = GridSearchCV(SVC(), param_grid, cv=5, return_train_score=True) grid_search.fit(X_train, y_train) print("최적 파라미터: {}".format(grid_search.best_params_)) print("최고 교차 검증 점수: {:.2f}".format(grid_search.best_score_)) # grid_search.cv_results_ pd.DataFrame(grid_search.cv_results_).iloc[:, 4:].T.head() param_grid = {'C':[0.001, 0.01, 0.1, 1, 10, 100],'gamma':[0.001, 0.01, 0.1, 1, 10, 100]} gsc = GridSearchCV(SVC(), param_grid, cv=5, return_train_score=True) scores = cross_val_score(gsc, iris.data, iris.target, cv=5) print(scores, scores.mean())<jupyter_output>[0.96666667 1. 0.96666667 0.96666667 1. ] 0.9800000000000001
no_license
/머신러닝/l.evalueation(모델평가, 성능 향상).ipynb
chayoonjung/yun-python
1
<jupyter_start><jupyter_text>Create a array<jupyter_code>data = np.arange(10) print (data.shape) data data = data.reshape(2,5) print (data.shape) data data_1 = np.ones([3,3]) print (type(data_1)) print (data_1) data_1 = np.zeros([3,3]) print (data_1) data = np.eye(3,3) data data =np.diag((1,2,3)) data data = np.empty([3,3]) data data = np.random.rand(3,3) print (data) data = np.array([[1,2,3],[2,3,4],[3,4,5]]) data data_list = list(range(9)) data = np.array(data_list).reshape(3,3) data<jupyter_output><empty_output><jupyter_text>Matrix Operation<jupyter_code>a = np.arange(3) print (a) a.shape b = a.T print (b) b.shape a = a.reshape(1,3) # row first, column second print (a) a.shape b = a.T print (b) b.shape c = np.dot(b,a) c c = b @ a c c = c + np.ones([3,3]) c c ** 2 d = c * c d a = np.random.rand(3,3) a all_max = a.max() x_max = a.max(axis = 0) # axis means along which axis y_max = a.max(axis = 1) print (all_max,x_max,y_max) all_max_arg = a.argmax() x_max_arg = a.argmax(axis = 0) print (all_max_arg, x_max_arg) a = np.random.rand(12) a ind = a.argsort() ind a.sort() a<jupyter_output><empty_output><jupyter_text>Indexing, Slicing and Iterating<jupyter_code>b= a[:,2] b for row in a: print (row) for element in a.flat: print (element) a = np.random.rand(4,4) a b = a.reshape(8,-1) b<jupyter_output><empty_output><jupyter_text>Stacking together different arrays<jupyter_code>a = np.random.rand(2,2) b = np.random.rand(2,2) print (a,b) np.vstack((a,b)) # vertical stack np.hstack((a,b)) # horizontal stack<jupyter_output><empty_output><jupyter_text>Copies and Views<jupyter_code>a = np.arange(12) # No copy at all, different names, but share same memory address # shape of a and data are changable via b b= a # Shallow copy: shape of a doesn't change and data of a changes c = a.view() # Deep copy: a's shape and data don't change d= a.copy()<jupyter_output><empty_output><jupyter_text>Indexing with Boolean Arrays<jupyter_code>a = np.random.rand(3,3) a a[[0,2],[1,2]] a[a>0.5]<jupyter_output><empty_output><jupyter_text>Linear Algebra<jupyter_code>a = np.array([[1.0, 2.0], [3.0, 4.0]]) np.linalg.inv(a) np.linalg.eig(a) <jupyter_output><empty_output><jupyter_text>Questions<jupyter_code>data = np.eye(3,3) data np.nonzero(data) np.any(data)<jupyter_output><empty_output>
permissive
/Numpy.ipynb
yuzhipeter/Data_Structure_Numpy_Pandas
8
<jupyter_start><jupyter_text># Read in the data<jupyter_code>import pandas as pd import numpy import re data_files = [ "ap_2010.csv", "class_size.csv", "demographics.csv", "graduation.csv", "hs_directory.csv", "sat_results.csv" ] data = {} for f in data_files: d = pd.read_csv("schools/{0}".format(f)) data[f.replace(".csv", "")] = d<jupyter_output><empty_output><jupyter_text># Read in the surveys<jupyter_code>all_survey = pd.read_csv("schools/survey_all.txt", delimiter="\t", encoding='windows-1252') d75_survey = pd.read_csv("schools/survey_d75.txt", delimiter="\t", encoding='windows-1252') survey = pd.concat([all_survey, d75_survey], axis=0) survey["DBN"] = survey["dbn"] survey_fields = [ "DBN", "rr_s", "rr_t", "rr_p", "N_s", "N_t", "N_p", "saf_p_11", "com_p_11", "eng_p_11", "aca_p_11", "saf_t_11", "com_t_11", "eng_t_11", "aca_t_11", "saf_s_11", "com_s_11", "eng_s_11", "aca_s_11", "saf_tot_11", "com_tot_11", "eng_tot_11", "aca_tot_11", ] survey = survey.loc[:,survey_fields] data["survey"] = survey<jupyter_output><empty_output><jupyter_text># Add DBN columns<jupyter_code>data["hs_directory"]["DBN"] = data["hs_directory"]["dbn"] def pad_csd(num): string_representation = str(num) if len(string_representation) > 1: return string_representation else: return "0" + string_representation data["class_size"]["padded_csd"] = data["class_size"]["CSD"].apply(pad_csd) data["class_size"]["DBN"] = data["class_size"]["padded_csd"] + data["class_size"]["SCHOOL CODE"]<jupyter_output><empty_output><jupyter_text># Convert columns to numeric<jupyter_code>cols = ['SAT Math Avg. Score', 'SAT Critical Reading Avg. Score', 'SAT Writing Avg. Score'] for c in cols: data["sat_results"][c] = pd.to_numeric(data["sat_results"][c], errors="coerce") data['sat_results']['sat_score'] = data['sat_results'][cols[0]] + data['sat_results'][cols[1]] + data['sat_results'][cols[2]] def find_lat(loc): coords = re.findall("\(.+, .+\)", loc) lat = coords[0].split(",")[0].replace("(", "") return lat def find_lon(loc): coords = re.findall("\(.+, .+\)", loc) lon = coords[0].split(",")[1].replace(")", "").strip() return lon data["hs_directory"]["lat"] = data["hs_directory"]["Location 1"].apply(find_lat) data["hs_directory"]["lon"] = data["hs_directory"]["Location 1"].apply(find_lon) data["hs_directory"]["lat"] = pd.to_numeric(data["hs_directory"]["lat"], errors="coerce") data["hs_directory"]["lon"] = pd.to_numeric(data["hs_directory"]["lon"], errors="coerce")<jupyter_output><empty_output><jupyter_text># Condense datasets<jupyter_code>class_size = data["class_size"] class_size = class_size[class_size["GRADE "] == "09-12"] class_size = class_size[class_size["PROGRAM TYPE"] == "GEN ED"] class_size = class_size.groupby("DBN").agg(numpy.mean) class_size.reset_index(inplace=True) data["class_size"] = class_size data["demographics"] = data["demographics"][data["demographics"]["schoolyear"] == 20112012] data["graduation"] = data["graduation"][data["graduation"]["Cohort"] == "2006"] data["graduation"] = data["graduation"][data["graduation"]["Demographic"] == "Total Cohort"]<jupyter_output><empty_output><jupyter_text># Convert AP scores to numeric<jupyter_code>cols = ['AP Test Takers ', 'Total Exams Taken', 'Number of Exams with scores 3 4 or 5'] for col in cols: data["ap_2010"][col] = pd.to_numeric(data["ap_2010"][col], errors="coerce")<jupyter_output><empty_output><jupyter_text># Combine the datasets<jupyter_code>combined = data["sat_results"] combined = combined.merge(data["ap_2010"], on="DBN", how="left") combined = combined.merge(data["graduation"], on="DBN", how="left") to_merge = ["class_size", "demographics", "survey", "hs_directory"] for m in to_merge: combined = combined.merge(data[m], on="DBN", how="inner") combined = combined.fillna(combined.mean()) combined = combined.fillna(0)<jupyter_output><empty_output><jupyter_text># Add a school district column for mapping<jupyter_code>def get_first_two_chars(dbn): return dbn[0:2] combined["school_dist"] = combined["DBN"].apply(get_first_two_chars)<jupyter_output><empty_output><jupyter_text># Find correlations<jupyter_code>correlations = combined.corr() correlations = correlations["sat_score"] print(correlations)<jupyter_output>SAT Critical Reading Avg. Score 0.986820 SAT Math Avg. Score 0.972643 SAT Writing Avg. Score 0.987771 sat_score 1.000000 AP Test Takers 0.523140 Total Exams Taken 0.514333 Number of Exams with scores 3 4 or 5 0.463245 Total Cohort 0.325144 CSD 0.042948 NUMBER OF STUDENTS / SEATS FILLED 0.394626 NUMBER OF SECTIONS 0.362673 AVERAGE CLASS SIZE 0.381014 SIZE OF SMALLEST CLASS 0.249949 SIZE OF LARGEST CLASS 0.314434 SCHOOLWIDE PUPIL-TEACHER RATIO NaN schoolyear NaN fl_percent NaN frl_percent -0.722225 total_enrollment 0.367857 ell_num -0.153778 ell_percent [...]<jupyter_text># Plotting survey correlations<jupyter_code># Remove DBN since it's a unique identifier, not a useful numerical value for correlation. survey_fields.remove("DBN") %matplotlib inline combined.corr()["sat_score"][survey_fields].plot.bar() combined.plot.scatter(x="saf_s_11", y="sat_score") import matplotlib.pyplot as plt from mpl_toolkits.basemap import Basemap districts = combined.groupby("school_dist").agg(numpy.mean) districts.reset_index(inplace=True) m = Basemap( projection='merc', llcrnrlat=40.496044, urcrnrlat=40.915256, llcrnrlon=-74.255735, urcrnrlon=-73.700272, resolution='i' ) m.drawmapboundary(fill_color='#85A6D9') m.drawcoastlines(color='#6D5F47', linewidth=.4) m.drawrivers(color='#6D5F47', linewidth=.4) longitudes = districts["lon"].tolist() latitudes = districts["lat"].tolist() m.scatter(longitudes, latitudes, s=50, zorder=2, latlon=True, c=districts["saf_s_11"], cmap="summer") plt.show() race_fields = ["white_per", "asian_per", "black_per", "hispanic_per"] combined.corr()["sat_score"][race_fields].plot.bar() combined.plot.scatter(x="hispanic_per", y="sat_score") print(combined[combined["hispanic_per"] > 95]["SCHOOL NAME"]) low_hispanic = combined[combined["hispanic_per"] < 10] print(low_hispanic[low_hispanic["sat_score"] > 1800]["SCHOOL NAME"]) print(combined[(combined["hispanic_per"] < 10) & (combined["sat_score"] > 1800)]["SCHOOL NAME"]) gender = ["male_per", "female_per"] print(combined.corr()["sat_score"][gender].plot.bar()) print(combined.plot.scatter(x="female_per", y="sat_score")) print(combined[(combined["female_per"] > 60) & (combined["sat_score"] > 170)]["SCHOOL NAME"]) combined["ap_per"] = combined["AP Test Takers "]/combined["total_enrollment"] print(combined.plot.scatter(x="ap_per", y="sat_score"))<jupyter_output>Axes(0.125,0.125;0.775x0.775)
no_license
/Guided Project: Analyzing NYC High School Data/Schools.ipynb
isabellechiu/Self-Project-Dataquest
10
<jupyter_start><jupyter_text># Mô tả dữ liệu bằng Arviz ### Bs. Lê Ngọc Khả Nhi Bài thực hành nảy nhằm hướng dẫn dùng package Arviz để thực hiện các biểu đồ mô tả đữ liệu đơn giản. Arviz (https://arviz-devs.github.io/arviz/index.html) là một thư viện đồ họa chuyên dụng cho Thống kê Bayes, cho phép vẽ các biểu đồ để diễn giải posterior distribution từ các mô hình Bayes. Tuy nhiên, ta có thể lợi dụng các hàm của Arviz cho mục tiêu thống kê mô tả đơn giản, vì sẽ tiện lợi hơn nhiều so với vẽ thủ công bằng matplotlib.<jupyter_code>import arviz as az import numpy as np import pandas as pd print('Arviz', az.__version__)<jupyter_output>Arviz 0.6.1 <jupyter_text>Trước khi dùng arviz, dữ liệu đầu vào cần được chuyển đổi thành cấu trúc "inference_data" bằng hàm convert_to_inference_data; hàm này chấp nhận những output từ mô hình Bayes (pymc3, pystan, pyro, ...); tuy nhiên nó cũng dung nạp 2 kiểu đữ liệu không phải là mô hình Bayes, gồm numpy array và dictionary## Từ 1D numpy array<jupyter_code>n = 1000 np_1D_dat = az.convert_to_inference_data(np.random.normal(2,2,size = n)) print(np_1D_dat) np_1D_dat.posterior<jupyter_output>Inference data with groups: > posterior <jupyter_text>## Từ pandas Series<jupyter_code>pd_S = pd.Series(np.random.gamma(0.5, size = n)) pd_S_dat = az.convert_to_inference_data(pd_S.values) print(pd_S_dat) pd_S_dat.posterior<jupyter_output>Inference data with groups: > posterior <jupyter_text>## Từ nD numpy array (ma trận 2D, tensor, array đa chiều)<jupyter_code>k = (1,1000,3) np_3D_dat = az.convert_to_inference_data(np.random.randn(*k)) print(np_3D_dat) np_3D_dat.posterior<jupyter_output>Inference data with groups: > posterior <jupyter_text>## Từ Dictionary<jupyter_code>datadict = { 'X': np.random.normal(1,2,size = n), 'Y': np.random.normal(1,2,size = n)*2 + np.random.rand(1000), 'Z': np.random.normal(1,2,size = n)*(-2) + np.random.rand(1000), } data_dict = az.convert_to_inference_data(datadict) data_dict.posterior<jupyter_output><empty_output><jupyter_text>## Từ pandas dataframe<jupyter_code>pd_df = pd.DataFrame(datadict) pd_df.head() data_df= az.convert_to_inference_data(pd_df.to_numpy()) data_df.posterior data_df= az.convert_to_inference_data(pd_df.values) data_df.posterior<jupyter_output><empty_output><jupyter_text>Có nhiều phong cách mỹ thuật khác nhau để chọn:<jupyter_code>az.style.available az.style.use('arviz-whitegrid')<jupyter_output><empty_output><jupyter_text># Hàm plot_dist Hàm này cho phép biểu diễn hàm mật độ phân phối (PDF, CDF), biểu đồ 1D KDE, 1D Histogram và 2D KDE Ưu điểm : arviz tự động nhận ra biến liên tục và dùng KDE, biến rời rạc để dùng histogram. Nó cho phép kết hợp thêm Rugs layer, hỗ trợ hàm CDF lẫn PDF<jupyter_code>az.plot_dist(np_1D_dat.posterior.x, rug = True, color ="red", fill_kwargs={'alpha': 0.3, 'color':'#eb3483'},) az.plot_dist(np_1D_dat.posterior.x, rug = True, cumulative =True, color ="red", fill_kwargs={'alpha': 0.3, 'color':'#eb3483'},) az.plot_dist(np.random.poisson(5, 1000), color ="#eb3483",) az.plot_dist(np.random.poisson(5, 1000), color ="#eb3483", fill_kwargs={'alpha': 0.3, 'color':'#eb3483'}, kind = 'kde')<jupyter_output><empty_output><jupyter_text>Hơn thế nữa, hàm plot_dist vẽ được 2D KDE rất đẹp:<jupyter_code>az.plot_dist(np_1D_dat.posterior.x, np_1D_dat.posterior.x * 2 + np.random.normal(0,5,1000), contour = True, contour_kwargs={'colors': 'white','alpha':0.8,} ) az.plot_dist(np_1D_dat.posterior.x, np_1D_dat.posterior.x * 2 + np.random.normal(0,5,1000), contour = False, contour_kwargs={'colors': 'white','alpha':0.8,} )<jupyter_output><empty_output><jupyter_text>## Hàm plot_posterior: Suy diễn thống kê Hàm này dùng để suy diễn phân phối hậu nghiệm của các mô hình Bayes, tuy nhiên không gì ngăn cản chúng ta dùng nó cho dữ liệu 1D bất kì. Hàm này dùng làm thống kê mô tả cực kì hiệu quả, vì nó cho phép kết hợp 1D KDE plot và các trị số thống kê median, mean, mode, khoảng tin cậy, và suy diễn thống kê dựa vào các method Bayes như HDP (tương tự như CI), Khoảng vô nghĩa thực dụng (ROPE), ngưỡng vô hiệu (ref_val):<jupyter_code>az.plot_posterior(np_1D_dat, var_names = ['x'], credible_interval = 0.95, point_estimate = 'median', rope = (-1.5,7), ref_val = 0, **{'alpha': 0.8, 'color':'blue'}) az.plot_posterior(np_1D_dat, var_names = ['x'], credible_interval = 0.95, point_estimate = 'median', rope = (-1.5,7), ref_val = 0, kind = 'hist',)<jupyter_output><empty_output><jupyter_text>Hàm này dùng tốt cho dữ liệu đa biến:<jupyter_code>az.plot_posterior(np_3D_dat)<jupyter_output><empty_output><jupyter_text>## Hàm plot_density Hàm plot_density có công dụng đơn giản hơn plot_posterior, nó trình bày thêm 1 điểm giá trị Median hoặc mean, mode; Ưu thế của hàm này đó là cho phép so sánh 2 hay nhiều phân nhóm:<jupyter_code>np_3D_dat2 = az.convert_to_inference_data(np.random.randn(*k)-1) az.plot_density(np_3D_dat, figsize = (12,3), shade = 0.3, point_estimate = "median", credible_interval =1, colors = "red") az.plot_density([np_3D_dat, np_3D_dat2], figsize = (12,3), shade = 0.3, credible_interval =0.95, point_estimate = "median")<jupyter_output><empty_output><jupyter_text>## Hàm plot_joint Hàm plot_joint cho phép biểu diễn hình ảnh Xác suất kết hợp (Joint pribability) và Xác suất biên (Marginal prob) trên cùng 1 biểu đồ. Xác suất kết hợp có thể biểu diễn bằng 3 kiểu hình họa: tán xạ, hex_bin và 2D KDE Xác suất biên có 2 kiểu hình họa: 1D KDE và Histogram<jupyter_code>datadict = { 'X': np.random.normal(2,2,size = n), 'Y': np.random.normal(2,2,size = n)* 2 + np.random.normal(0,5,1000), } az.plot_joint(datadict) az.plot_joint(datadict, joint_kwargs = {'color':"red", 'alpha': 0.3}, marginal_kwargs = {'color':'red', 'fill_kwargs':{'alpha': 0.3, 'color':'#eb3483'}} ) az.plot_joint(datadict, figsize=(5.7,5), kind = 'hexbin', marginal_kwargs = {'color':'black', 'fill_kwargs':{'alpha': 0.9, 'color':'#ebbd34'}} ) az.plot_joint(datadict, figsize=(5.7,5), kind = 'kde', marginal_kwargs = {'color':'black', 'fill_kwargs':{'alpha': 0.9, 'color':'#ebbd34'}} )<jupyter_output><empty_output><jupyter_text>## Hàm plot_pair Hàm này là sự mở rộng của hàm plot_join, vì nó cho phép khảo sát bắt cặp tuần tự dữ liệu có từ 3 biến trở lên (ma trận tương quan).<jupyter_code>datadict = { 'X': np.random.normal(1,2,size = n), 'Y': np.random.normal(1,2,size = n)*2 + np.random.rand(1000), 'Z': np.random.normal(1,2,size = n)*(-2) + np.random.rand(1000), } az.plot_pair(datadict) az.plot_pair(np_3D_dat, figsize=(5.7,5), coords = {'x_dim_0': [0, 1, 2]}, var_names = ['x'], kind = 'kde', divergences = True, ) az.plot_pair(datadict, figsize=(5.7,5), colorbar=True, kind ='hexbin')<jupyter_output><empty_output><jupyter_text>## Hàm plot_forest Hàm plot_forest cho phép biểu diễn nội dung các posterior distribution của coeficients các mô hình Bayes; tuy nhiên ta có thể dùng nó để so sánh trực quan giữa nhiều phân nhóm hay nhiều biến có cùng thang đo (tương đương ridge_plot, forest_plot)<jupyter_code>datadict = { 'X': np.random.normal(1,2,size = n), 'Y': np.random.normal(1,2,size = n)*2 + np.random.rand(1000), 'Z': np.random.normal(1,2,size = n)*(-2) + np.random.rand(1000), } az.plot_forest(datadict, kind="forestplot", rope = (-5,5), credible_interval = 0.95, colors = "red") az.plot_forest(datadict, kind="ridgeplot", credible_interval = 0.95, ridgeplot_alpha = 0.5, ridgeplot_overlap = 3, colors = "red")<jupyter_output><empty_output>
no_license
/Arviz demo 1.ipynb
kinokoberuji/Statistics-Python-Tutorials
15
<jupyter_start><jupyter_text># Programming and Data Analysis > Homework 0 Kuo, Yao-Jen from [DATAINPOINT](https://www.datainpoint.com)## Instructions - We've imported necessary modules at the beginning of each exercise. - We've put necessary files(if any) in the working directory of each exercise. - We've defined the names of functions/inputs/parameters for you. - Write down your solution between the comments `### BEGIN SOLUTION` and `### END SOLUTION`. - It is NECESSARY to `return` the answer, tests will fail by just printing out the answer. - Do not use `input()` function, it will halt the notebook while running tests. - Running tests to see if your solutions are right: Kernel -> Restart & Run All -> Restart and Run All Cells. - You can run tests after each question or after finishing all questions.<jupyter_code>import unittest<jupyter_output><empty_output><jupyter_text>## 01. Define a function named `convert_fahrenheit_to_celsius(x)` which converts Fahrenheit degrees to Celsius degrees. \begin{equation} Celsius^{\circ} C = (Fahrenheit^{\circ} F - 32) \times \frac{5}{9} \end{equation} - Expected inputs:a numeric `x`. - Expected outputs:a numeric.<jupyter_code>def convert_fahrenheit_to_celsius(x): """ >>> convert_fahrenheit_to_celsius(212) 100.0 >>> convert_fahrenheit_to_celsius(32) 0.0 """ ### BEGIN SOLUTION y = (x - 32) * 5/9 return y ### END SOLUTION<jupyter_output><empty_output><jupyter_text>## 02. Define a function named `calculate_bmi(height, weight)` which calculates BMI according to heights in meters and weights in kilograms. \begin{equation} BMI = \frac{weight_{kg}}{height_{m}^2} \end{equation} Source: - Expected inputs:2 numerics `height` and `weight`. - Expected outputs:a numeric.<jupyter_code>def calculate_bmi(height, weight): """ >>> calculate_bmi(216, 147) # Shaquille O'Neal in his prime 31.507201646090532 >>> calculate_bmi(206, 113) # LeBron James 26.628334433028563 >>> calculate_bmi(211, 110) # Giannis Antetokounmpo 24.70744143213315 """ ### BEGIN SOLUTION BMI = weight / ((0.01 * height) * (0.01 * height)) return BMI ### END SOLUTION<jupyter_output><empty_output><jupyter_text>## 03. Define a function named `show_big_mac_index(country, currency, price)` which returns the Big Mac Index given a country, its currency, and the price of a Big Mac. - Expected inputs:2 strings and a numeric. - Expected outputs:a string.<jupyter_code>def show_big_mac_index(country, currency, price): """ >>> show_big_mac_index('US', 'USD', 5.65) A Big Mac costs 5.65 USD in US. >>> show_big_mac_index('South Korea', 'Won', 6520) A Big Mac costs 6,520.00 Won in South Korea. >>> show_big_mac_index('Taiwan', 'NTD', 72) A Big Mac costs 72.00 NTD in Taiwan. """ ### BEGIN SOLUTION ans = 'A Big Mac costs ' + str('{:0,.2f}'.format(price)) + ' ' + currency + ' in ' + country + '.' return ans ### END SOLUTION<jupyter_output><empty_output><jupyter_text>## 04. Define a function named `is_a_divisor(x, y)` which returns whether `x` is a is_a_divisor of `y` or not. - Expected inputs:2 integers. - Expected outputs:a boolean.<jupyter_code>def is_a_divisor(x, y): """ >>> is_a_divisor(1, 3) True >>> is_a_divisor(2, 3) False >>> is_a_divisor(3, 3) True >>> is_a_divisor(1, 4) True >>> is_a_divisor(2, 4) True >>> is_a_divisor(3, 4) False >>> is_a_divisor(4, 4) True """ ### BEGIN SOLUTION if y % x == 0: return True else: return False ### END SOLUTION<jupyter_output><empty_output><jupyter_text>## 05. Define a function named `contains_vowels(x)` which returns whether x contains one of the vowels: a, e, i, o, u or not. - Expected inputs:a string. - Expected outputs:a boolean.<jupyter_code>def contains_vowels(x): """ >>> contains_vowels('pythn') False >>> contains_vowels('ncnd') False >>> contains_vowels('rtclt') False >>> contains_vowels('python') True >>> contains_vowels('anaconda') True >>> contains_vowels('reticulate') True """ ### BEGIN SOLUTION if ('a' in x) or ('e' in x) or ('i' in x) or ('o' in x) or ('u' in x): return True else: return False ### END SOLUTION<jupyter_output><empty_output><jupyter_text>## Run tests! Kernel -> Restart & Run All. -> Restart And Run All Cells.<jupyter_code>class TestHomeworkZero(unittest.TestCase): def test_01_convert_fahrenheit_to_celsius(self): self.assertAlmostEqual(convert_fahrenheit_to_celsius(212), 100.0) self.assertAlmostEqual(convert_fahrenheit_to_celsius(32), 0.0) def test_02_calculate_bmi(self): self.assertTrue(calculate_bmi(216, 147) > 31) self.assertTrue(calculate_bmi(216, 147) < 32) self.assertTrue(calculate_bmi(206, 113) > 26) self.assertTrue(calculate_bmi(206, 113) < 27) self.assertTrue(calculate_bmi(211, 110) > 24) self.assertTrue(calculate_bmi(211, 110) < 25) def test_03_show_big_mac_index(self): self.assertEqual(show_big_mac_index('US', 'USD', 5.65), 'A Big Mac costs 5.65 USD in US.') self.assertEqual(show_big_mac_index('South Korea', 'Won', 6520), 'A Big Mac costs 6,520.00 Won in South Korea.') self.assertEqual(show_big_mac_index('Taiwan', 'NTD', 72), 'A Big Mac costs 72.00 NTD in Taiwan.') def test_04_is_a_divisor(self): self.assertTrue(is_a_divisor(1, 2)) self.assertTrue(is_a_divisor(2, 2)) self.assertTrue(is_a_divisor(1, 3)) self.assertFalse(is_a_divisor(2, 3)) self.assertTrue(is_a_divisor(1, 4)) self.assertTrue(is_a_divisor(2, 4)) self.assertFalse(is_a_divisor(3, 4)) self.assertTrue(is_a_divisor(4, 4)) def test_05_contains_vowels(self): self.assertFalse(contains_vowels('pythn')) self.assertFalse(contains_vowels('ncnd')) self.assertFalse(contains_vowels('rtclt')) self.assertTrue(contains_vowels('python')) self.assertTrue(contains_vowels('anaconda')) self.assertTrue(contains_vowels('reticulate')) suite = unittest.TestLoader().loadTestsFromTestCase(TestHomeworkZero) runner = unittest.TextTestRunner(verbosity=2) test_results = runner.run(suite) number_of_failures = len(test_results.failures) number_of_errors = len(test_results.errors) number_of_test_runs = test_results.testsRun number_of_successes = number_of_test_runs - (number_of_failures + number_of_errors) print("You've got {} successes among {} questions.".format(number_of_successes, number_of_test_runs))<jupyter_output>You've got 5 successes among 5 questions.
no_license
/exercises.ipynb
rose020/homework0
7
<jupyter_start><jupyter_text>#### Plotting of gesture data<jupyter_code>print("Shape of X_train: ", X_train.shape) print("shape of y_train/labels: ", y_train.shape) print("Shape of X_test: ", X_test.shape) print("shape of y_test/labels: ", y_test.shape) samples = np.random.choice(len(X_train), 8) def show_images(images, cols = 1, titles = None): """Display a list of images in a single figure with matplotlib. Parameters --------- images: List of np.arrays compatible with plt.imshow. cols (Default = 1): Number of columns in figure (number of rows is set to np.ceil(n_images/float(cols))). titles: List of titles corresponding to each image. Must have the same length as titles. """ assert((titles is None)or (len(images) == len(titles))) n_images = len(images) if titles is None: titles = ['Image (%d)' % i for i in range(1,n_images + 1)] fig = plt.figure() for n, (image, title) in enumerate(zip(images, titles)): a = fig.add_subplot(cols, np.ceil(n_images/float(cols)), n + 1) if image.ndim == 2: plt.gray() plt.imshow(image) a.set_title(title, fontsize=50) a.grid(False) a.axis("off") fig.set_size_inches(np.array(fig.get_size_inches()) * n_images) plt.show() sample_images = [] sample_labels = [] for sample in samples: sample_images.append(X_train[sample]) for key, val in label_dict.items(): if np.argmax(y_train[sample]) == val: sample_labels.append(key) show_images(sample_images, 2, titles=sample_labels)<jupyter_output><empty_output><jupyter_text>### Model<jupyter_code>def create_model(): model = Sequential() model.add(Conv2D(64, kernel_size = [3,3], padding = 'same', activation = 'relu', input_shape = (200,200,3))) model.add(Conv2D(64, kernel_size = [3,3], padding = 'same', activation = 'relu')) model.add(MaxPool2D(pool_size = [3,3])) model.add(Conv2D(128, kernel_size = [5,5], padding = 'same', activation = 'relu')) model.add(Conv2D(128, kernel_size = [5,5], padding = 'same', activation = 'relu')) model.add(MaxPool2D(pool_size = [3,3])) model.add(Conv2D(256, kernel_size = [3,3], padding = 'same', activation = 'relu')) model.add(Conv2D(256, kernel_size = [3,3], padding = 'same', activation = 'relu')) model.add(Conv2D(256, kernel_size = [3,3], padding = 'same', activation = 'relu')) model.add(MaxPool2D(pool_size = [3,3])) model.add(Conv2D(512, kernel_size = [3,3], padding = 'same', activation = 'relu')) model.add(Conv2D(512, kernel_size = [3,3], padding = 'same', activation = 'relu')) model.add(MaxPool2D(pool_size = [3,3])) model.add(Conv2D(512, kernel_size = [3,3], padding = 'same', activation = 'relu')) model.add(MaxPool2D(pool_size = [2,2])) model.add(BatchNormalization()) model.add(Flatten()) model.add(Dropout(0.5)) model.add(Dense(1024, activation = 'relu', kernel_regularizer = regularizers.l2(0.001))) model.add(Dense(512, activation = 'relu', kernel_regularizer = regularizers.l2(0.001))) model.add(Dense(36, activation = 'softmax')) print("MODEL CREATED") return model model = create_model() model.summary() model.compile(optimizer = 'adam', loss = keras.losses.categorical_crossentropy, metrics = ["accuracy"]) model_hist = model.fit(X_train, y_train, batch_size = 32, epochs = 15, validation_data=(X_test, y_test))<jupyter_output>WARNING:tensorflow:From /opt/conda/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version. Instructions for updating: Use tf.cast instead. Train on 24026 samples, validate on 1265 samples Epoch 1/15 24026/24026 [==============================] - 84s 3ms/step - loss: 2.1800 - acc: 0.5566 - val_loss: 2.0808 - val_acc: 0.5407 Epoch 2/15 24026/24026 [==============================] - 81s 3ms/step - loss: 0.8187 - acc: 0.8212 - val_loss: 0.5425 - val_acc: 0.8838 Epoch 3/15 24026/24026 [==============================] - 81s 3ms/step - loss: 0.4958 - acc: 0.8852 - val_loss: 0.8439 - val_acc: 0.7739 Epoch 4/15 24026/24026 [==============================] - 81s 3ms/step - loss: 0.3759 - acc: 0.9093 - val_loss: 0.3088 - val_acc: 0.9241 Epoch 5/15 24026/24026 [==============================] - 81s 3ms/step - loss: 0.3117 - acc: 0.9276 - val_loss: 0.3129 - val_acc: 0.9273 Epoch 6/15 24[...]<jupyter_text>### Plotting training metrices<jupyter_code>def plot_accuracy(y): if(y == True): plt.plot(model_hist.history['acc']) plt.plot(model_hist.history['val_acc']) plt.legend(['train', 'test'], loc='lower right') plt.title('accuracy plot - train vs test') plt.xlabel('epoch') plt.ylabel('accuracy') plt.show() else: pass return def plot_loss(y): if(y == True): plt.plot(model_hist.history['loss']) plt.plot(model_hist.history['val_loss']) plt.legend(['training loss', 'validation loss'], loc = 'upper right') plt.title('loss plot - training vs vaidation') plt.xlabel('epoch') plt.ylabel('loss') plt.show() else: pass return plot_accuracy(True) plot_loss(True) model.save("asl_bestsofar.h5") samples_test = np.random.choice(len(X_test), 8) samples_test sample_images = [] sample_labels = [] pred_labels = [] for sample in samples_test: sample_images.append(X_test[sample]) img = X_test[sample].reshape((1,200,200,3)) pred = model.predict_classes(img) for key, val in label_dict.items(): if pred[0] == int(val): pred_labels.append(key) for key, val in label_dict.items(): if np.argmax(y_test[sample]) == val: sample_labels.append(key) def show_test_images(images, cols = 1, true_label = None, pred_label=None): n_images = len(images) fig = plt.figure() for n, (image, label, pred) in enumerate(zip(images, true_label, pred_label)): a = fig.add_subplot(cols, np.ceil(n_images/float(cols)), n + 1) if image.ndim == 2: plt.gray() plt.imshow(image) a.set_title("{}\n{}".format(label, pred), fontsize=50) a.grid(False) a.axis("off") fig.set_size_inches(np.array(fig.get_size_inches()) * n_images) plt.show() show_test_images(sample_images, 2, sample_labels, pred_labels)<jupyter_output><empty_output>
no_license
/asl-training.ipynb
ayulockin/ASL_Classifier
3
<jupyter_start><jupyter_text># Pandas basics In this notebook we will **learn** how to work with the two main data types in `pandas`: `DataFrame` and `Series`.## Data structures (`pandas`)### `Series` In `pandas`, series are the building blocks of dataframes. Think of a series as a column in a table. A series collects *observations* about a given *variable*. <jupyter_code>from random import random import pandas as pd import numpy as np from pandas import Series, DataFrame<jupyter_output><empty_output><jupyter_text>#### Numerical series<jupyter_code># let's create a series containing 100 random numbers # ranging between 0 and 1 s = pd.Series([random() for n in range(0, 10)])<jupyter_output><empty_output><jupyter_text>Each observation in the series has an **index** as well as a set of **values**: they can be accessed via the omonymous properties:<jupyter_code>s.index list(s.index) s.values<jupyter_output><empty_output><jupyter_text>The `head()` and `tail()` methods allows for looking at the begininning and end of a series:<jupyter_code>s.head() s.tail()<jupyter_output><empty_output><jupyter_text>The `value_counts()` method returns a count of distinct values within a series.Is there any number in `s` that occurs twice?<jupyter_code># a `Series` can be easily cast into a list list(s.value_counts()).count(2)<jupyter_output><empty_output><jupyter_text>Another way of verifying this:<jupyter_code>s.is_unique s.min() s.max() s.mean() s.median()<jupyter_output><empty_output><jupyter_text>#### Datetime series<jupyter_code>from random import randint from datetime import date # let's generate a list of random dates # in the range 1900-1950 dates = [ date( year, randint(1, 12), randint(1, 28) # try replacing with 31 and see what happens ) for year in range(1900,1950) ] s1 = pd.Series(dates) s1 type(s1[1]) s1 = Series(pd.to_datetime(dates)) type(s1[1]) s1[1].day_name() s1.min() s1.max() s1.mean()<jupyter_output><empty_output><jupyter_text>### `DataFrame` What is a `pandas.DataFrame`? Think of it as an in-memory spreadsheet that you can analyse and manipulate programmatically. A `DataFrame` is a collection of `Series` having the same length and whose indexes are in sync. A *collection* means that each column of a dataframe is a seriesLet's create a toy `DataFrame` by hand. <jupyter_code>dates = [ date( year, randint(1, 12), randint(1, 28) # try replacing with 31 and see what happens ) for year in range(1980,1990) ] counts = [ randint(0, 10000) for i in range(0, 10) ] event_types = ["fire", "flood", "car_crash", "plane_crash"] events = [ np.random.choice(event_types) for i in range(0, 10) ] assert len(events) == len(counts) == len(dates) toy_df = pd.DataFrame({ "date": dates, "count": counts, "event": events }) toy_df<jupyter_output><empty_output><jupyter_text>**Try out**: what happens if you change the length of either of the two lists? Try e.g. passing 20 dates instead of 10.<jupyter_code># a df is a collection of series # each column is a series type(toy_df.date)<jupyter_output><empty_output><jupyter_text>## Data manipulation in `pandas`### Data types String, datetimes (see above), categorical data. In `pandas`, categories behave very much like string, yet they lead to better performances (faster operations, optimized storage).Bottom-up approach:<jupyter_code># transforms a Series with strings into categories toy_df.event.astype('category')<jupyter_output><empty_output><jupyter_text>### Exploring a dataframe Exploring a dataframe: df.head(), df.tail(), df.info().The method `info()` gives you information about a dataframe: - how much space does it take in memory? - what is the datatype of each column? - how many records are there? - how many `null` values does each column contain (!)?<jupyter_code>toy_df.info()<jupyter_output><class 'pandas.core.frame.DataFrame'> RangeIndex: 10 entries, 0 to 9 Data columns (total 3 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 date 10 non-null object 1 count 10 non-null int64 2 event 10 non-null object dtypes: int64(1), object(2) memory usage: 368.0+ bytes <jupyter_text>Alternatively, if you need to know only the number of columns and rows you can use the `.shape` property. It returns a tuple with 1) number of rows, 2) number of columns.<jupyter_code>toy_df.shape<jupyter_output><empty_output><jupyter_text>`head()` prints by first five rows of a dataframe:<jupyter_code>toy_df.head()<jupyter_output><empty_output><jupyter_text>But the number of lines displayed is a parameter that can be changed:<jupyter_code>toy_df.head(2)<jupyter_output><empty_output><jupyter_text>`tail()` does the opposite, i.e. prints the last n rows in the dataframe:<jupyter_code>toy_df.tail()<jupyter_output><empty_output><jupyter_text>#### Adding columnsLet's go back to our toy dataframe:<jupyter_code>toy_df.head()<jupyter_output><empty_output><jupyter_text>Using the column selector with the name of a column that does not exist yet will add the effect of setting the values of all rows in that column to the value specified.<jupyter_code>toy_df['country'] = "UK" toy_df.head(3)<jupyter_output><empty_output><jupyter_text>But if the column already exists, its value is reset:<jupyter_code>toy_df['country'] = "USA" toy_df.head(3)<jupyter_output><empty_output><jupyter_text>#### Removing columnsThe double square bracket notation ``[[...]]`` returns a dataframe having only the columns specified inside the inner brackets.This said, removing a column is done by unselecting it:<jupyter_code># here we removed the column country toy_df2 = toy_df[['date', 'count', 'event']] # it worked! toy_df2.head()<jupyter_output><empty_output><jupyter_text>#### Setting a column as index<jupyter_code>toy_df.set_index('date') toy_df.head(3) toy_df.set_index('date', inplace=True) toy_df.head(3)<jupyter_output><empty_output><jupyter_text>**Q**: can you explain the effect of the `inplace` parameter by looking at the cells above?### Accessing data .loc, .iloc, slicing, iteration over rows<jupyter_code>toy_df.head(3)<jupyter_output><empty_output><jupyter_text>#### Label-based indexing<jupyter_code>toy_df.loc[date(1980,1,1):date(1982,1,1)]<jupyter_output><empty_output><jupyter_text>#### Integer-based indexing<jupyter_code># select a single row, the first one toy_df.iloc[0] # select a range of rows by index toy_df.iloc[[1,3,-1]] # select a range of rows with slicing toy_df.iloc[0:5] toy_df.index<jupyter_output><empty_output><jupyter_text>#### Iterating over rows<jupyter_code>for n, row in toy_df.iterrows(): print(n) for n, row in toy_df.iterrows(): print(n, row.event)<jupyter_output>1980-01-23 plane_crash 1981-04-15 plane_crash 1982-11-12 flood 1983-11-22 plane_crash 1984-11-04 fire 1985-06-19 flood 1986-09-14 plane_crash 1987-01-05 fire 1988-03-11 flood 1989-10-10 car_crash
permissive
/notebooks/3_PandasBasics.ipynb
Giovanni1085/UvA_CDH_2020
24
<jupyter_start><jupyter_text><jupyter_code>import numpy as np # useful for many scientific computing in Python import pandas as pd # primary data structure library !conda install -c anaconda xlrd --yes df_can = pd.read_excel('https://s3-api.us-geo.objectstorage.softlayer.net/cf-courses-data/CognitiveClass/DV0101EN/labs/Data_Files/Canada.xlsx', sheet_name='Canada by Citizenship', skiprows=range(20), skipfooter=2) print ('Data read into a pandas dataframe!') #Dataset of immigration into canada from other countries - 1980 to 2013 (each year and each country) df_can.head(10) df_can.shape df_can.tail() df_can.info() #simplify and remove unnecessary columns df_can.drop(['AREA','REG','DEV','Type','Coverage'], inplace=True, axis='columns') df_can.head() #rename columns to sensible names df_can.rename(columns={'OdName':'Country', 'AreaName':'Continent', 'RegName':'Continent-Region'}, inplace=True) df_can.head() #count total immigration by country for all years df_can['total_immigration'] = df_can.sum(axis='columns') df_can.head() df_can.describe() df_can.isnull().sum() df_can.dtypes #Select immigration of years 1980-1985 from the dataset, for all countries? #get the list of columns in the dataframe df_can.columns df_can[['Country',1980,1981,1982,1983,1984,1985]] #simplify the dataset by changing the index from numbers to country names df_can.index.values df_can.set_index('Country', inplace=True) df_can.head() print(df_can.loc['India']) df_can.iloc[8] print(df_can.loc['Japan',2000]) print(df_can.iloc[12,12]) #convert columns names to string #map - syntax(function, collection) #apply function to each element of the collection - equivalent apply, mapply,sapply,lapply functions of R df_can.columns = list(map(str,df_can.columns)) df_can.columns #filtering on dataframe #Q. find all countries in Asia condition = df_can['Continent'] == 'Asia' df_can[condition] #Q. multiple conditions df_can[(df_can['Continent']=='Asia')] df_can[(df_can['1980'] > 1000)] #syntax - df[(condition)] df_can[(df_can['Continent']=='Asia')&(df_can['2013']>10000)] df_can.where(df_can['Continent']=='Asia') #conditions #basic stats import matplotlib.pyplot as plt import matplotlib as mpl import plotly as px mpl.__version__ df_can.columns years = list(map(str,range(1980,2014))) years india = df_can.loc['India',years] india india.plot() india.index #change the index to integer india.index = india.index.map(int) india.plot(kind='line') plt.title('Immigration from India') plt.ylabel('Number of immigrants') plt.xlabel('Years') #update the plot plt.show() india.plot(kind='line') plt.title('Immigration from India') plt.ylabel('Number of immigrants') plt.xlabel('Years') plt.text(2000, 32000, 'Y2K Revolution') plt.text(1990,15000,'Recession') plt.text(1985,10000,'Economic boom') #update the plot plt.show() years = list(map(str,range(1980,2014))) china = df_can.loc['China', years] china.plot() china.index = china.index.map(int) china.plot(kind='line') plt.title('Immigration of China') plt.xlabel('Years') plt.ylabel('Immigrants') plt.text(1990,10000,'Marker') plt.show() years = list(map(str,range(1980,2014))) chinaindia = df_can.loc[['China','India'], years] new_chinaindia = chinaindia.transpose() new_chinaindia.plot() new_chinaindia.index = new_chinaindia.index.map(int) new_chinaindia.plot(kind='line') plt.title('Immigration of China and India') plt.xlabel('Years') plt.ylabel('Immigrants') #plt.text(1990,10000,'Marker') plt.show() #Q. Which two countries have similar immigration trends over the years 1980-2013? #Q. France and Germany #top 5 countries that send immigrants to canada df_can.sort_values(by='total_immigration', ascending=False, axis='index', inplace=True) top5 = df_can.head(5) top5 years = list(map(str,range(1980,2014))) top5_clean = top5[years] top5_clean = top5_clean.transpose() top5_clean.plot()<jupyter_output><empty_output><jupyter_text>#Part 2<jupyter_code>#Area plots - Stacked line plot top5_clean top5_clean.index = top5_clean.index.map(int) top5_clean.plot(kind="area", stacked=False, figsize=(20,10)) plt.title('Immigration trends in top five countries (1980-2013)') plt.xlabel('Years') plt.ylabel('Immigrants count') plt.show() top5_clean.index = top5_clean.index.map(int) top5_clean.plot(kind="area", stacked=True, figsize=(20,10)) plt.title('Immigration trends in top five countries (1980-2013)') plt.xlabel('Years') plt.ylabel('Immigrants count') plt.show() top5_clean.index = top5_clean.index.map(int) #alpha parameter for transparency - default = 0.5 (range is 0 to 1) top5_clean.plot(kind="area", stacked=False, figsize=(20,10), alpha=0.25) plt.title('Immigration trends in top five countries (1980-2013)') plt.xlabel('Years') plt.ylabel('Immigrants count') plt.show() #Bottom five countries - stacked and unstacked area plot #histogram - at a particular time period / snapshot of the data df_can['2000'] df_can['2000'].plot(kind='hist', figsize=(20,10)) plt.title('Immigration trends in 195 countries in 2000') plt.xlabel('Number of Immigrants') plt.ylabel('Number of Countries') plt.show() #histogram years = list(map(str,range(1980,2014))) df_can.loc[['India','China','Denmark','Norway','France','Germany'], years].transpose().plot.hist() df_can.loc[['India','China','Denmark','Norway','France','Germany'], years].transpose().plot.hist() plt.title('Immigration trends in 6 countries in 1980-2013') plt.xlabel('Number of Immigrants') plt.ylabel('Number of Countries') plt.show()<jupyter_output><empty_output><jupyter_text>#Vertical bar plot<jupyter_code>india = df_can.loc['India', years] india.plot(kind='bar', figsize=(20,10)) plt.title('Immigration trends India immigrants from 1980 to 2013') plt.xlabel('Years') plt.ylabel('India immigrants from 1980 to 2013') plt.show() india = df_can.loc['India', years] india.index = india.index.map(int) india.plot(kind='bar', figsize=(10,6)) plt.title('Immigration trends India immigrants from 1980 to 2013') plt.xlabel('Years') plt.ylabel('India immigrants from 1980 to 2013') plt.annotate('Increasing trends', xy=(32, 70), xycoords='data', xytext=(28, 20), arrowprops=dict(facecolor='black', shrink=0.05), horizontalalignment='left', verticalalignment='top', ) plt.show() india = df_can.loc['India', years] india.plot(kind='barh', figsize=(20,10)) plt.title('Immigration trends India immigrants from 1980 to 2013') plt.xlabel('India immigrants from 1980 to 2013') plt.ylabel('Years') plt.show()<jupyter_output><empty_output>
permissive
/Lab Experiments/Experiment_3_200720.ipynb
rohitsmittal7/J045-ML-Sem-V
3
<jupyter_start><jupyter_text>**MATH 3332 ** **Section 52 ** # In Who-is-Normal.xslx there are 7 columns representing variables x1-x7, one of which is sample from the normal distribution. Find which one of the variables is normal? ## Reading in Values The program starts by reading in the values from the Excel spreadsheet after they have been exported to a CSV format.<jupyter_code>import csv f = open('input.csv', 'r') reader = csv.reader(f) values = [] for row in reader: values.append(row)<jupyter_output><empty_output><jupyter_text>## Manipulating the Values The program then transposes the matrix that was created when reading in the values from the CSV. The result is an array, where the rows consist of the name of the data (e.g., x1, x2) and the values for that distribution. <jupyter_code>import numpy as np matrix = np.transpose(np.array(values))<jupyter_output><empty_output><jupyter_text>## Creating the Probability Plot The program then creates a plot for each distribution using the **scipy.stats** library. The plot contains the **plotted values**, the $R^2$ value, the **mean**, and the **standard deviation **.<jupyter_code>import scipy.stats import matplotlib.pyplot as plt import matplotlib.gridspec as grid %matplotlib inline # Iterate over each dataset (x1, x2, etc) for x_column in xrange(len(matrix)): # Create array of float-type values from the dataset values = np.array(matrix[x_column][1:]) values = [float(value) for value in values] # Create a seperate figure to create the probability plot fig = plt.figure() fig.text(0, 0, 'Mean %s' % np.mean(values)) fig.text(1, 0, 'Std %s' % np.std(values)) fig.suptitle('x%s' % (x_column + 1)) # Produce the probability plot scipy.stats.probplot(values, plot=plt) # Create a separate figure to create the histogram fig2 = plt.figure() fig2.suptitle('x%s Histogram' % (x_column + 1)) # Produce the histogram using normalized values plt.hist(values, normed=True) # Add the normal distribution curve to the plot mu, std = scipy.stats.norm.fit(values) xmin, xmax = plt.xlim() x = np.linspace(xmin, xmax, 100) p = scipy.stats.norm.pdf(x, mu, std) plt.plot(x, p, 'k', linewidth=2) plt.show()<jupyter_output><empty_output>
no_license
/Probability/Normal.ipynb
rlacherksu/notebooks
3
<jupyter_start><jupyter_text># 株価の関係<jupyter_code>%matplotlib inline import matplotlib.pyplot as plt import matplotlib import numpy import sqlite3 import pandas as pd import os fmdate="2015-01-01" todate="2016-12-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = os.environ["HOME"]+'/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") sql="""select date as tm,close from histDaily where code=? and date between ? and ? order by tm""" res=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) dt=pd.to_datetime(res["tm"].values) plt.scatter(dt,res["close"].values) plt.xticks(rotation=90) plt.title("株価ヒストリカル", fontproperties=fp) plt.xlabel("date") plt.ylabel("price") plt.show()<jupyter_output><empty_output><jupyter_text>## 出来高との関係 出来高が増える。注目度が高くなる。なので<jupyter_code>%matplotlib inline import matplotlib.pyplot as plt import numpy import sqlite3 import pandas as pd import matplotlib fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = r'/Users/admin/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/stock/daily.db") sql="""select strftime('%Y%m%d',date) as tm,volume from histDaily where code=? and date between ? and ? order by tm""" res=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) dt=pd.to_datetime(res["tm"].values) plt.scatter(dt,res["volume"].values) plt.xticks(rotation=90) plt.title("出来高のヒストリカル", fontproperties=fp) plt.xlabel("date") plt.ylabel("volume") plt.show()<jupyter_output><empty_output><jupyter_text>## 投稿文字スコアの推移 1or-1に修正し合算する,悲観的な投稿が多いか少ないかを見る<jupyter_code>%matplotlib inline import matplotlib.pyplot as plt import numpy as np import sqlite3 import pandas as pd fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = r'/Users/admin/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") # https://stackoverflow.com/questions/41324503/pandas-sqlite-query-using-variable sql="""select strftime('%Y%m%d',b.date) as tm,sum(case when e.score >=0 then 1 else -1 end) as score from board b ,board_score e where b.code=e.code and b.tno=e.tno and b.mno=e.mno and b.code=? and b.date between ? and ? group by tm order by tm""" res=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) #https://qiita.com/ColdFreak/items/1028927c81fdfa5f3ac2 dt=pd.to_datetime(res["tm"].values) plt.scatter(dt,res["score"].values) plt.xticks(rotation=90) plt.title("投稿スコアの推移", fontproperties=fp) plt.xlabel("tm") plt.ylabel("log10(avg(score))") plt.show()<jupyter_output><empty_output><jupyter_text>## ここまでの予想 - 出来高が上がると注目度が高くなる - そうすると人が増える - 怪しい情報も増えてくる ### 絵文字は置いといて -時系列で集まる人数をカウントする。## 書き込み人数を時系列で出力 <jupyter_code># emoji kind %matplotlib inline import matplotlib import matplotlib.pyplot as plt import numpy import sqlite3 import os import pandas as pd fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = r'/Users/admin/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") cur=con.cursor() res=pd.read_sql_query("select tm,count(*) as user_count from (select strftime('%Y%m%d',date) as tm,user from board"+ " where date between ? and ? and code=? group by tm,user) group by tm",con,params=(fmdate,todate,code,)) dt=pd.to_datetime(res["tm"].values) plt.scatter(dt,res["user_count"].values) plt.xticks(rotation=90) plt.title("投稿者数と時系列", fontproperties=fp) plt.xlabel("tm") plt.ylabel("user_count") plt.show() <jupyter_output><empty_output><jupyter_text>## 相関を見る - 投稿者数と出来高の相関係数 - 営業日以外の場合もあるし,投稿のない場合もあるのでInnerJoin<jupyter_code># emoji kind %matplotlib inline import matplotlib import matplotlib.pyplot as plt import numpy as np import sqlite3 import os import pandas as pd from sklearn import linear_model fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = r'/Users/admin/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") cur=con.cursor() user=pd.read_sql_query("select tm,count(*) as user_count from (select strftime('%Y%m%d',date) as tm,user from board"+ " where date between ? and ? and code=? group by tm,user) group by tm",con,params=(fmdate,todate,code,)) user["tm"]=pd.to_datetime(user["tm"]) #user=user.set_index("tm") #cur.close() #con.close() print(user.head()) #con=sqlite3.connect("../../../../data/stock/daily.db") #cur=con.cursor() volume=pd.read_sql_query("select strftime('%Y%m%d',date) as tm,volume/10000 as volume from histDaily where code=? and date between ? and ? order by tm",con,params=(code,fmdate,todate,)) volume["tm"]=pd.to_datetime(volume["tm"]) #volume=volume.set_index("tm") cur.close() con.close() print(volume.head()) data=pd.merge(volume,user,how="inner",left_on="tm",right_on="tm") print(data.head()) # https://qiita.com/m-hayashi/items/ee379c86e3e18f0ddc6d # http://pythondatascience.plavox.info/scikit-learn/線形回帰 x=pd.DataFrame(data["volume"]).values y=pd.DataFrame(data["user_count"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("出来高と投稿者数の関係", fontproperties=fp) plt.xlabel("volume") plt.ylabel("user_count") plt.show()<jupyter_output> tm user_count 0 2015-01-01 1 1 2015-01-02 2 2 2015-01-03 1 3 2015-01-04 1 4 2015-01-05 1 tm volume 0 2015-01-05 2110.3 1 2015-01-06 3103.5 2 2015-01-07 2395.0 3 2015-01-08 1928.2 4 2015-01-09 3816.0 tm volume user_count 0 2015-01-05 2110.3 1 1 2015-01-06 3103.5 2 2 2015-01-07 2395.0 6 3 2015-01-08 1928.2 1 4 2015-01-09 3816.0 11 [[ 0.02094571]] [-17.38032316] 0.709948491945 <jupyter_text>- 出来高が増えると投稿者数も増える - 要は出来高が増えると目立つので参加者が増えるということ## 投稿スコアと出来高の関係 どうやら出来高で説明ができるのか?<jupyter_code># emoji kind %matplotlib inline import matplotlib import matplotlib.pyplot as plt import numpy as np import sqlite3 import os import pandas as pd from sklearn import linear_model fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = r'/Users/admin/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") cur=con.cursor() sql="""select strftime('%Y%m%d',b.date) as tm,sum(case when e.score >=0 then 1 else -1 end) as score,avg(e.score/e.wordnum) as ascore from board b ,board_score e where b.code=e.code and b.tno=e.tno and b.mno=e.mno and b.code=? and b.date between ? and ? group by tm order by tm""" score=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) #https://qiita.com/ColdFreak/items/1028927c81fdfa5f3ac2 score["tm"]=pd.to_datetime(score["tm"]) #cur.close() #con.close() print(score.head()) #con=sqlite3.connect("../../../../data/stock/daily.db") #cur=con.cursor() volume=pd.read_sql_query("select strftime('%Y%m%d',date) as tm,volume/10000 as volume from histDaily where code=? and date between ? and ? order by tm",con,params=(code,fmdate,todate,)) volume["tm"]=pd.to_datetime(volume["tm"]) #volume=volume.set_index("tm") cur.close() con.close() print(volume.head()) data=pd.merge(volume,score,how="inner",left_on="tm",right_on="tm") print(data.head()) # https://qiita.com/m-hayashi/items/ee379c86e3e18f0ddc6d # http://pythondatascience.plavox.info/scikit-learn/線形回帰 x=pd.DataFrame(data["volume"]).values y=pd.DataFrame(data["score"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("出来高と投稿スコアの関係,スコアは-1to1に修正", fontproperties=fp) plt.xlabel("volume") plt.ylabel("score") plt.show() ## 平均化 x=pd.DataFrame(data["volume"]).values y=pd.DataFrame(data["ascore"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("出来高と投稿スコアの関係,スコアはscore/wordnumに修正", fontproperties=fp) plt.xlabel("volume") plt.ylabel("score") plt.show() <jupyter_output> tm score ascore 0 2015-01-01 1 0.043008 1 2015-01-02 -3 -0.463773 2 2015-01-03 -1 -0.399109 3 2015-01-04 -1 -0.455821 4 2015-01-05 -1 -0.555359 tm volume 0 2015-01-05 2110.3 1 2015-01-06 3103.5 2 2015-01-07 2395.0 3 2015-01-08 1928.2 4 2015-01-09 3816.0 tm volume score ascore 0 2015-01-05 2110.3 -1 -0.555359 1 2015-01-06 3103.5 -2 -0.728097 2 2015-01-07 2395.0 -7 -0.479000 3 2015-01-08 1928.2 -1 -0.605556 4 2015-01-09 3816.0 -13 -0.503629 [[-0.09497118]] [ 101.33264451] 0.650158690959 <jupyter_text>出来高と投稿スコアにも相関関係がある## 投稿スコアと参加者数の関係 となるとこれも相関があるのか?<jupyter_code># emoji kind %matplotlib inline import matplotlib import matplotlib.pyplot as plt import numpy as np import sqlite3 import os import pandas as pd from sklearn import linear_model fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = os.environ["HOME"]+'/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") cur=con.cursor() sql="""select strftime('%Y%m%d',b.date) as tm,sum(case when e.score >=0 then 1 else -1 end) as score,avg(e.score/e.wordnum) as ascore from board b ,board_score e where b.code=e.code and b.tno=e.tno and b.mno=e.mno and b.code=? and b.date between ? and ? group by tm order by tm""" score=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) #https://qiita.com/ColdFreak/items/1028927c81fdfa5f3ac2 score["tm"]=pd.to_datetime(score["tm"]) user=pd.read_sql_query("select tm,count(*) as user_count from (select strftime('%Y%m%d',date) as tm,user from board"+ " where date between ? and ? and code=? group by tm,user) group by tm",con,params=(fmdate,todate,code,)) user["tm"]=pd.to_datetime(user["tm"]) #user=user.set_index("tm") cur.close() con.close() print(user.head()) print(score.head()) data=pd.merge(user,score,how="inner",left_on="tm",right_on="tm") print(data.head()) # https://qiita.com/m-hayashi/items/ee379c86e3e18f0ddc6d # http://pythondatascience.plavox.info/scikit-learn/線形回帰 x=pd.DataFrame(data["user_count"]).values y=pd.DataFrame(data["score"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("参加者数と投稿スコアの関係,スコア-1to1修正", fontproperties=fp) plt.xlabel("user_count") plt.ylabel("score") plt.show() #### x=pd.DataFrame(data["user_count"]).values y=pd.DataFrame(data["ascore"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("参加者数と投稿スコアの関係,スコアscore/wordnum", fontproperties=fp) plt.xlabel("user_count") plt.ylabel("score") plt.show()<jupyter_output> tm user_count 0 2015-01-01 1 1 2015-01-02 2 2 2015-01-03 1 3 2015-01-04 1 4 2015-01-05 1 tm score ascore 0 2015-01-01 1 0.043008 1 2015-01-02 -3 -0.463773 2 2015-01-03 -1 -0.399109 3 2015-01-04 -1 -0.455821 4 2015-01-05 -1 -0.555359 tm user_count score ascore 0 2015-01-01 1 1 0.043008 1 2015-01-02 2 -3 -0.463773 2 2015-01-03 1 -1 -0.399109 3 2015-01-04 1 -1 -0.455821 4 2015-01-05 1 -1 -0.555359 [[-4.60171765]] [ 28.35209875] 0.948489885375 <jupyter_text>どうやら人が増えるとスコアが下がる。下げたい人間が多いようだ## 参加者が多い時,少ない時の絵文字の特徴 - 参加者が多い時は下げようとする傾向が高いので絵文字がスコア低くなる? - 1日ごとの絵文字別投稿数と参加者数,もしくは出来高との相関 - 使われている絵文字の上位10個 これはそのまま出して,一投稿1絵文字種類に修正する。(同じ絵文字を繰り返し使う投稿の影響削除)<jupyter_code>%matplotlib inline import matplotlib import matplotlib.pyplot as plt import numpy as np import sqlite3 import os import pandas as pd from sklearn import linear_model fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = os.environ["HOME"]+'/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") # top 10 sql="select c.emoji,sum(c.num),s.score from board_emoji_count c,emoji_score s,board b where b.code=c.code and b.tno=c.tno and b.mno=c.mno and c.emoji=s.emoji and c.code=? and b.date between ? and ? group by c.emoji order by sum(c.num) desc limit 10" emoji_list=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) print(emoji_list) for em in emoji_list["emoji"]: print(em) user=pd.read_sql_query("select tm,count(*) as user_count from (select strftime('%Y%m%d',date) as tm,user from board"+ " where date between ? and ? and code=? group by tm,user) group by tm",con,params=(fmdate,todate,code,)) user["tm"]=pd.to_datetime(user["tm"]) emoji=pd.read_sql_query("select strftime('%Y%m%d',date) as tm,sum(c.num) as emoji_count from board b ,board_emoji_count c"+ " where b.code=c.code and b.tno=c.tno and b.mno=c.mno and b.date between ? and ? and b.code=? and c.emoji=? group by tm",con,params=(fmdate,todate,code,em,)) emoji["tm"]=pd.to_datetime(emoji["tm"]) data=pd.merge(user,emoji,how="inner") data.drop("tm",axis=1) x=pd.DataFrame(data["user_count"]).values y=pd.DataFrame(data["emoji_count"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("絵文字"+em+"数とユーザ数の関係", fontproperties=fp) plt.xlabel("user_count") plt.ylabel("emoji_score") plt.show() con.close() <jupyter_output> emoji sum(c.num) score 0 🙌 3329 1.00 1 😱 908 -0.75 2 💛 549 0.75 3 💀 514 -0.50 4 💩 450 -0.75 5 😁 383 0.50 6 😅 381 0.50 7 👍 180 0.50 8 😊 165 0.50 9 😃 134 0.50 🙌 [[ 0.06752109]] [ 33.30399642] 0.0204427406372 <jupyter_text>## 一投稿人絵文字にDistinctする<jupyter_code>%matplotlib inline import matplotlib import matplotlib.pyplot as plt import numpy as np import sqlite3 import os import pandas as pd from sklearn import linear_model fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = os.environ["HOME"]+'/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") # top 10 sql="select c.emoji,sum(c.num),s.score from board_emoji_count c,emoji_score s,board b where b.code=c.code and b.tno=c.tno and b.mno=c.mno and c.emoji=s.emoji and c.code=? and b.date between ? and ? group by c.emoji order by sum(c.num) desc limit 10" emoji_list=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) print(emoji_list) for em in emoji_list["emoji"]: print(em) user=pd.read_sql_query("select tm,count(*) as user_count from "+ "(select strftime('%Y%m%d',date) as tm,user from board "+ " where date between ? and ? and code=? group by tm,user) "+ "group by tm",con,params=(fmdate,todate,code,)) user["tm"]=pd.to_datetime(user["tm"]) emoji=pd.read_sql_query("select strftime('%Y%m%d',date) as tm,count(*) as emoji_count "+ " from board b ,board_emoji_count c"+ " where b.code=c.code and b.tno=c.tno and b.mno=c.mno and "+ " b.date between ? and ? and b.code=? and c.emoji=? group by tm",con,params=(fmdate,todate,code,em,)) emoji["tm"]=pd.to_datetime(emoji["tm"]) data=pd.merge(user,emoji,how="inner") data.drop("tm",axis=1) x=pd.DataFrame(data["user_count"]).values y=pd.DataFrame(data["emoji_count"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("絵文字"+em+"数とユーザ数の関係", fontproperties=fp) plt.xlabel("user_count") plt.ylabel("emoji_score") plt.show() con.close() <jupyter_output> emoji sum(c.num) score 0 🙌 3329 1.00 1 😱 908 -0.75 2 💛 549 0.75 3 💀 514 -0.50 4 💩 450 -0.75 5 😁 383 0.50 6 😅 381 0.50 7 👍 180 0.50 8 😊 165 0.50 9 😃 134 0.50 🙌 [[ 0.00145165]] [ 0.92381301] 0.23822882925 <jupyter_text># 絵文字投稿率と出来高 絵文字の投稿率と出来高は関係があるか?出来高が増えると人が増える,絵文字率が増えるか。1日あたりの投稿率で<jupyter_code>%matplotlib inline import matplotlib import matplotlib.pyplot as plt import numpy as np import sqlite3 import os import pandas as pd from sklearn import linear_model fmdate="2015-01-01" todate="2017-07-31" code="6502" if os.name == "nt": fname=r'C:\WINDOWS\Fonts\ipaexg.ttf' else: fname = os.environ["HOME"]+'/Library/Fonts/ipaexg.ttf' fp = matplotlib.font_manager.FontProperties(fname=fname, size=14) con=sqlite3.connect(os.environ["DATA"]+"/board/board.db") # emoji post sql="select strftime('%Y%m%d',b.date) as tm,count(*) emoji_post_count from board b,board_emoji_count e where b.code=e.code and b.mno=e.mno and b.tno=e.mno and e.code=? and b.date between ? and ? group by tm order by tm " emoji_list=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) # all post sql="select strftime('%Y%m%d',date) as tm,count(*) all_post_count from board where code=? and date between ? and ? group by tm" allpost_list=pd.read_sql_query(sql,con,params=(code,fmdate,todate,)) data=pd.merge(emoji_list,allpost_list,how="inner",left_on="tm",right_on="tm") data["emoji_post_ratio"]=data["emoji_post_count"]/data["all_post_count"] con.close() # 出来高 con=sqlite3.connect("../../../../data/stock/daily.db") cur=con.cursor() volume=pd.read_sql_query("select strftime('%Y%m%d',date) as tm,volume/10000 as volume from histDaily where code=? and date between ? and ? order by tm",con,params=(code,fmdate,todate,)) #volume["tm"]=pd.to_datetime(volume["tm"]) #volume=volume.set_index("tm") cur.close() con.close() data=pd.merge(data,volume,how="inner",left_on="tm",right_on="tm") x=pd.DataFrame(data["volume"]).values y=pd.DataFrame(data["emoji_post_ratio"]).values model=linear_model.LinearRegression() model.fit(x,y) # 回帰係数 print(model.coef_) # 切片 (誤差) print(model.intercept_) # 決定係数 print(model.score(x, y)) plt.scatter(x,y,color="red") plt.plot(x,model.predict(x)) plt.title("絵文字投稿率と出来高", fontproperties=fp) plt.xlabel("volume") plt.ylabel("emoji_post_ratio") plt.show()<jupyter_output>[[ -4.82129903e-07]] [ 0.01603249] 0.12813828662
no_license
/theme/notebook/22_toshiba_stock.ipynb
k-utsubo/xc40
10
<jupyter_start><jupyter_text>## Online Factorization Machine Online factorization models take single data as an input, make a prediction, and train with the data.### 1. Setup The from models imports the package for use. We have also imported a few other packages for plotting.<jupyter_code>import sys sys.path.append('./../') from utils import data_preprocess, plot import os import pickle import numpy as np import torch from time import time from models.models_online.FM_FTRL import FM_FTRL from models.models_online.SFTRL_CCFM import SFTRL_CCFM from models.models_online.SFTRL_Vanila import SFTRL_Vanila from models.models_online.RRF_Online import RRF_Online from utils.data_manager import * from utils.metric_manager import * Tensor_type = torch.DoubleTensor import seaborn as sns import os sns.set() sns.set_style('white') import matplotlib plt.rcParams["axes.grid"] = True plt.rc('font', family='serif') current_palette = sns.color_palette(sns.hls_palette(5, l=.6, s=1.0)) sns.palplot(current_palette) current_palette = np.asarray(current_palette)<jupyter_output><empty_output><jupyter_text>### 2. Create a dataset depending on task We prepare movielens100-k for regression and cod-ran dataset for classification<jupyter_code>task = 'cls' if task == 'reg': nbUsers = 943 nbMovies = 1682 nbFeatures = nbUsers + nbMovies nbRatingsTrain = 90570 nbRatingsTest = 9430 data_dir = os.getcwd() + '/../dataset/ml-100k/' #filename1, filename2 = 'ub.base', 'ub.test' filename1, filename2 = 'ua.base', 'ua.test' _, x_train, y_train, rate_train, timestamp_train = load_dataset_movielens(data_dir + filename1,nbRatingsTrain,nbFeatures,nbUsers) x_train_s, rate_train_s, _ = sort_dataset_movielens(x_train, rate_train, timestamp_train) want_permute = True if want_permute : idx = np.random.permutation(x_train_s.shape[0]) x_train_s = x_train_s[idx] rate_train_s = rate_train_s[idx] else: pass x_train_s = x_train_s.todense() elif task == 'cls': data_dir = './../dataset/cod-rna2/' filename = "cod-rna2.scale" X, y = load_svmlight_file(data_dir + filename, n_features=8) X = X.toarray() want_permute = False if want_permute : idx = np.random.permutation(X.shape[0]) x_train_s = np.asarray(X[idx]) rate_train_s = np.asarray(y[idx]) else: x_train_s = np.asarray(X) rate_train_s = np.asarray(y) else: raise ValueError<jupyter_output><empty_output><jupyter_text>### 3. Create factorization machines and conduct online task There are four models in this notebook; 'FTRL','SFTRL_C','SFTRL_V','Online_RFF'<jupyter_code>model_option_list = ['FTRL','SFTRL_C','SFTRL_V','Online_RFF'] model_option = 'SFTRL_C' #model_option = 'Online_RFF' m = 40 lr_FM = 0.005 assert(model_option in model_option_list) if model_option is 'FTRL': Model_FM_FTRL = FM_FTRL(Tensor_type(x_train_s),Tensor_type(rate_train_s) , task,lr_FM, m ) pred_y, label_y , time = Model_FM_FTRL.online_learning() elif model_option is 'SFTRL_C': Model_SFTRL_C = SFTRL_CCFM(Tensor_type(x_train_s),Tensor_type(rate_train_s) , task,lr_FM,m ) pred_y, label_y , time = Model_SFTRL_C.online_learning() elif model_option is 'SFTRL_V': Model_SFTRL_V = SFTRL_Vanila(Tensor_type(x_train_s),Tensor_type(rate_train_s) , task,lr_FM,m ) pred_y, label_y , time = Model_SFTRL_V.online_learning() elif model_option in 'Online_RFF': inputs_matrix, outputs = Tensor_type(x_train_s), Tensor_type(rate_train_s) Online_RRF = RRF_Online(inputs_matrix,outputs,task) pred_y, label_y , time = Online_RRF.online_learning() #print(type(x_train_s)) <jupyter_output>======================================== SFTRL_CCFM_0.005_40_start 0 th : pred 1.000000 , real -1.000000 1000 th : pred -1.000000 , real -1.000000 2000 th : pred -1.000000 , real 1.000000 3000 th : pred -1.000000 , real -1.000000 4000 th : pred -1.000000 , real -1.000000 5000 th : pred -1.000000 , real -1.000000 6000 th : pred -1.000000 , real -1.000000 7000 th : pred -1.000000 , real -1.000000 8000 th : pred -1.000000 , real -1.000000 9000 th : pred -1.000000 , real -1.000000 10000 th : pred -1.000000 , real -1.000000 11000 th : pred -1.000000 , real -1.000000 12000 th : pred -1.000000 , real -1.000000 13000 th : pred -1.000000 , real -1.000000 14000 th : pred -1.000000 , real -1.000000 15000 th : pred -1.000000 , real -1.000000 16000 th : pred -1.000000 , real -1.000000 17000 th : pred -1.000000 , real -1.000000 18000 th : pred -1.000000 , real -1.000000 19000 th : pred -1.000000 , real -1.000000 20000 th : pred -1.000000 , real -1.000000 21[...]<jupyter_text>### 4. Cacluate Metric We prepare the each metric for regression (average root mean squre) and classification (average accuracy )<jupyter_code>if task == 'reg': metric = regression_metric(pred_y,label_y) elif task == 'cls': _,metric = classfication_metric(pred_y,label_y)<jupyter_output><empty_output><jupyter_text>### 5. Results figures We shows the resutls of prediction and accuracy for regression and classification<jupyter_code>fontsiz = 15 fig = plt.figure(figsize=(8*2,5.5)) fig.add_subplot(1,2,1) plt.plot(pred_y,'b.', label = 'pred',alpha = 0.8) plt.plot(label_y,'r.' , label = 'label',alpha = 0.8) plt.xlabel('iteration' , fontsize = fontsiz) plt.ylabel('y' , fontsize = fontsiz) plt.title('pred vs real' ,fontsize = fontsiz) plt.tick_params(axis='both', labelsize=fontsiz) #plt.legend(loc='upper center', bbox_to_anchor=(0, 1. + 0.075 ,1-0.05,0.1), ncol = 2 , fontsize = fontsiz) plt.legend( fontsize = fontsiz) fig.add_subplot(1,2,2) plt.plot(metric,'b', label = 'metric',alpha = 0.8 ,linewidth = 2.0) plt.xlabel('iteration' , fontsize = fontsiz) plt.ylabel('accumulated metric' , fontsize = fontsiz) plt.title('accumulated metric over iteration' ,fontsize = fontsiz) plt.tick_params(axis='both', labelsize=fontsiz) #plt.legend(ncol = 2 , fontsize = 12) #plt.legend(loc='upper center', bbox_to_anchor=(0, 1. + 0.075 ,1-0.05,0.1), ncol = 2 , fontsize = fontsiz) plt.legend( fontsize = fontsiz) fig.tight_layout() <jupyter_output><empty_output>
no_license
/jupyters/online_models_example.ipynb
yejihan-dev/fm-for-online-recommendation
5
<jupyter_start><jupyter_text>### 运行一次来获得cookie - 注意填充自己的帐号密码<jupyter_code>import requests import time from selenium import webdriver def get_pixiv_cookie(pixiv_id,pixiv_pw): driver = webdriver.Chrome() # Optional argument, if not specified will search pat driver.get('https://accounts.pixiv.net/login'); time.sleep(0.5) account = driver.find_element_by_css_selector('input[autocomplete="username"]') account.send_keys(pixiv_id) time.sleep(10) password = driver.find_element_by_css_selector('input[autocomplete="current-password"]') password.send_keys(pixiv_pw) time.sleep(1) password.submit() time.sleep(10) cookie=driver.get_cookies() driver.close() return cookie f=open("account.txt") p_id = f.readline().rstrip() p_pw = f.readline().rstrip() f.close() cookies_list=get_pixiv_cookie(p_id,p_pw) f=open("cookie.txt","w") f.write(str(cookies_list)) f.close()<jupyter_output><empty_output><jupyter_text>### 下面的直接运行可以就可以根据id爬图了- 从保存的文件获取cookie<jupyter_code>f=open("cookie.txt","r") cookie_list=eval(f.readline()) f.close() s = requests.Session() for cookie in cookie_list: s.cookies.set(cookie['name'], cookie['value'])<jupyter_output><empty_output><jupyter_text>- 获取预载信息<jupyter_code>pid='86685722' delay=0.5 url="https://www.pixiv.net/artworks/"+pid hder1={ 'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/84.0.4147.135 Safari/537.36 Edg/84.0.522.63', } r=s.get(url,headers=hder1) soup=BeautifulSoup(r.text, 'html.parser') meta=soup.find('meta',id="meta-preload-data")# 格式化后可清晰看到这里是预载信息 js=json.loads(meta['content']) # print(json.dumps(js,indent=2))<jupyter_output><empty_output><jupyter_text>使用`print(soup.prettify())`打印输出之后,可以清楚的发现预载信息在一个叫`meta-preload-data`的`meta`中 之后将内容提取之后也格式化 这是json文件,使用`json.dumps(js,indent=2)`进行格式化输出 ![image.png](images/image2.png) js本身是列表,通过列表的形式获取需要的信息,这里需要的主要是url和页数信息- 获取图片<jupyter_code>def get_image(_url,_hder,_filename=None,_folder='img'): if _filename==None: img_path=urllib.parse.urlparse(_url).path _filename=img_path.split('/')[-1]#路径的最后一项是文件名 f=open(_folder+'/'+_filename,"wb") re=requests.get(_url,headers=_hder) f.write(re.content) f.close() hder2 = { 'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/84.0.4147.135 Safari/537.36 Edg/84.0.522.63', "referer":"https://www.pixiv.net/" } ori=js["illust"][pid]['urls']['original'] pagecount=js['illust'][pid]['pageCount'] for i in range(pagecount): url=ori.replace("p0","p"+str(i)) # print(url) get_image(url,hder2) time.sleep(delay)<jupyter_output><empty_output><jupyter_text>这里注意只有`pixiv.net`发出的原图请求`pximg.net`才会受理,这个需要在原图界面观察Network后筛选得到 ![image.png](images/image1.png)### 整合了以上代码<jupyter_code>import requests import urllib.request import urllib.parse from bs4 import BeautifulSoup from PIL import Image import sys import time import json def get_image(_url,_hder,_filename=None,_folder='img'): if _filename==None: img_path=urllib.parse.urlparse(_url).path _filename=img_path.split('/')[-1]#路径的最后一项是文件名 f=open(_folder+'/'+_filename,"wb") re=requests.get(_url,headers=_hder) f.write(re.content) f.close() def download_id(pid): f=open("cookie.txt","r") cookie_list=eval(f.readline()) f.close() s = requests.Session() for cookie in cookie_list: s.cookies.set(cookie['name'], cookie['value']) if(type(pid)==int): pid=str(pid) delay=0.5 url="https://www.pixiv.net/artworks/"+pid hder1={ 'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/84.0.4147.135 Safari/537.36 Edg/84.0.522.63', } r=s.get(url,headers=hder1) soup=BeautifulSoup(r.text, 'html.parser') meta=soup.find('meta',id="meta-preload-data") js=json.loads(meta['content']) hder2 = { 'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/84.0.4147.135 Safari/537.36 Edg/84.0.522.63', "referer":"https://www.pixiv.net/" } ori=js["illust"][pid]['urls']['original'] pagecount=js['illust'][pid]['pageCount'] for i in range(pagecount): url=ori.replace("p0","p"+str(i)) get_image(url,hder2) time.sleep(delay) download_id(90479236)<jupyter_output><empty_output><jupyter_text>### 尝试通过作者爬取作品id<jupyter_code>author_id='27691' url='https://www.pixiv.net/ajax/user/'+author_id+'/profile/all' hder={ 'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/84.0.4147.135 Safari/537.36 Edg/84.0.522.63', } js=json.loads(s.get(url,headers=hder).text) print(js["body"]["illusts"].keys())<jupyter_output><empty_output><jupyter_text>首先通过 ``` url='https://www.pixiv.net/users/'+author_id+'/artworks' soup=BeautifulSoup(s.get(url,headers=hder).text) print(soup.prettify()) ``` 直接爬取该页面信息,但是分析后发现没有id信息,然后打开浏览器加载这个页面,在Network下的XHR里可以找到传来id信息的页面,是https://www.pixiv.net/ajax/user/27691/profile/all ![image3.png](images/image3.png) 直接点击可以访问,说明不需要cookie,只需要浏览器UA即可. 得到所有插画的id就等同于得到所有图片了<jupyter_code>for id in list(js["body"]["illusts"].keys()): download_id(id)<jupyter_output><empty_output>
no_license
/pixiv.ipynb
Unknown-Chinese-User/pixiv-spider
7
<jupyter_start><jupyter_text>**This notebook is an exercise in the [Intro to Deep Learning](https://www.kaggle.com/learn/intro-to-deep-learning) course. You can reference the tutorial at [this link](https://www.kaggle.com/ryanholbrook/deep-neural-networks).** --- # Introduction # In the tutorial, we saw how to build deep neural networks by stacking layers inside a `Sequential` model. By adding an *activation function* after the hidden layers, we gave the network the ability to learn more complex (non-linear) relationships in the data. In these exercises, you'll build a neural network with several hidden layers and then explore some activation functions beyond ReLU. Run this next cell to set everything up!<jupyter_code>import tensorflow as tf # Setup plotting import matplotlib.pyplot as plt plt.style.use('seaborn-whitegrid') # Set Matplotlib defaults plt.rc('figure', autolayout=True) plt.rc('axes', labelweight='bold', labelsize='large', titleweight='bold', titlesize=18, titlepad=10) # Setup feedback system from learntools.core import binder binder.bind(globals()) from learntools.deep_learning_intro.ex2 import *<jupyter_output><empty_output><jupyter_text>In the *Concrete* dataset, your task is to predict the compressive strength of concrete manufactured according to various recipes. Run the next code cell without changes to load the dataset.<jupyter_code>import pandas as pd concrete = pd.read_csv('../input/dl-course-data/concrete.csv') concrete.head()<jupyter_output><empty_output><jupyter_text># 1) Input Shape # The target for this task is the column `'CompressiveStrength'`. The remaining columns are the features we'll use as inputs. What would be the input shape for this dataset?<jupyter_code>concrete.shape # YOUR CODE HERE input_shape = [8] # Check your answer q_1.check() # Lines below will give you a hint or solution code #q_1.hint() #q_1.solution()<jupyter_output><empty_output><jupyter_text># 2) Define a Model with Hidden Layers # Now create a model with three hidden layers, each having 512 units and the ReLU activation. Be sure to include an output layer of one unit and no activation, and also `input_shape` as an argument to the first layer.<jupyter_code>from tensorflow import keras from tensorflow.keras import layers # YOUR CODE HERE model = keras.Sequential([ layers.Dense(units= 512, activation= 'relu',input_shape= [8]), layers.Dense(units= 512, activation= 'relu'), layers.Dense(units= 512, activation= 'relu'), layers.Dense(units= 1) ]) # Check your answer q_2.check() # Lines below will give you a hint or solution code #q_2.hint() #q_2.solution()<jupyter_output><empty_output><jupyter_text># 3) Activation Layers # Let's explore activations functions some. The usual way of attaching an activation function to a `Dense` layer is to include it as part of the definition with the `activation` argument. Sometimes though you'll want to put some other layer between the `Dense` layer and its activation function. (We'll see an example of this in Lesson 5 with *batch normalization*.) In this case, we can define the activation in its own `Activation` layer, like so: ``` layers.Dense(units=8), layers.Activation('relu') ``` This is completely equivalent to the ordinary way: `layers.Dense(units=8, activation='relu')`. Rewrite the following model so that each activation is in its own `Activation` layer.<jupyter_code>### YOUR CODE HERE: rewrite this to use activation layers model = keras.Sequential([ layers.Dense(32, activation='relu', input_shape=[8]), layers.Dense(32, activation='relu'), layers.Dense(1), ]) model = keras.Sequential([ layers.Dense(units= 32, input_shape=[8]), layers.Activation('relu'), layers.Dense(units= 32), layers.Activation('relu'), layers.Dense(units= 1), ]) # Check your answer q_3.check() # Lines below will give you a hint or solution code #q_3.hint() #q_3.solution()<jupyter_output><empty_output><jupyter_text># Optional: Alternatives to ReLU # There is a whole family of variants of the `'relu'` activation -- `'elu'`, `'selu'`, and `'swish'`, among others -- all of which you can use in Keras. Sometimes one activation will perform better than another on a given task, so you could consider experimenting with activations as you develop a model. The ReLU activation tends to do well on most problems, so it's a good one to start with. Let's look at the graphs of some of these. Change the activation from `'relu'` to one of the others named above. Then run the cell to see the graph. (Check out the [documentation](https://www.tensorflow.org/api_docs/python/tf/keras/activations) for more ideas.)<jupyter_code># YOUR CODE HERE: Change 'relu' to 'elu', 'selu', 'swish'... or something else activation_layer = layers.Activation('swish') x = tf.linspace(-3.0, 3.0, 100) y = activation_layer(x) # once created, a layer is callable just like a function plt.figure(dpi=100) plt.plot(x, y) plt.xlim(-3, 3) plt.xlabel("Input") plt.ylabel("Output") plt.show()<jupyter_output><empty_output>
no_license
/Intro to Deep Learning/2 - Deep Neural Networks/exercise-deep-neural-networks.ipynb
mtamjidhossain/Kaggle-courses
6
<jupyter_start><jupyter_text># Answering Questions for the Chinook Record StoreThe Chinook record store has just signed a deal with a new record label, and you've been tasked with selecting the first three albums that will be added to the store, from a list of four. All four albums are by artists that don't have any tracks in the store right now - we have the artist names, and the genre of music they produce: Artist Name - Genre - Regal - Hip-Hop - Red Tone - Punk - Meteor and the Girls - Pop - Slim Jim Bites - Blues The record label specializes in artists from the USA, and they have given Chinook some money to advertise the new albums in the USA, so we're interested in finding out which genres sell the best in the USA.<jupyter_code>%%capture %load_ext sql %sql sqlite:///chinook.db<jupyter_output><empty_output><jupyter_text># Overview of the Data Query the database to get a list of all tables and views in our database:<jupyter_code>%%sql SELECT name, type FROM sqlite_master WHERE type IN ("table","view");<jupyter_output>Done. <jupyter_text># Selecting Albums to PurchaseWrite a query that returns each genre, with the number of tracks sold in the USA: - in absolute numbers - in percentages<jupyter_code>%%sql SELECT il.* FROM invoice_line il INNER JOIN invoice i ON i.invoice_id = il.invoice_id INNER JOIN customer c on c.customer_id = i.customer_id WHERE c.country = 'USA' LIMIT 10; %%sql WITH usa_sold AS(SELECT il.* FROM invoice_line il INNER JOIN invoice i ON i.invoice_id = il.invoice_id INNER JOIN customer c on c.customer_id = i.customer_id WHERE c.country = 'USA' ) SELECT g.name genre, count(us.invoice_line_id) tracks_sold FROM usa_sold us INNER JOIN track t ON t.track_id = us.track_id INNER JOIN genre g ON g.genre_id = t.genre_id GROUP BY 1 ORDER BY 2; %%sql WITH usa_sold AS(SELECT il.* FROM invoice_line il INNER JOIN invoice i ON i.invoice_id = il.invoice_id INNER JOIN customer c on c.customer_id = i.customer_id WHERE c.country = 'USA' ) SELECT g.name genre, count(us.invoice_line_id) tracks_sold, cast(count(us.invoice_line_id) AS FLOAT) / ( SELECT COUNT(*) from usa_sold ) percentage_sold FROM usa_sold us INNER JOIN track t ON t.track_id = us.track_id INNER JOIN genre g ON g.genre_id = t.genre_id GROUP BY 1 ORDER BY 2 DESC LIMIT 10; <jupyter_output>Done. <jupyter_text>Based on the table above and the four options mentioned at the beginning, the albums to choose would be from should be: - Red Tone - Punk - Slim Jim Bites - Blues - Meteor and the Girls - Pop<jupyter_code>%%sql WITH usa_sold AS(SELECT il.* FROM invoice_line il INNER JOIN invoice i ON i.invoice_id = il.invoice_id INNER JOIN customer c on c.customer_id = i.customer_id WHERE c.country = 'USA' ) SELECT DISTINCT t.composer artist_name, g.name genre, count(us.invoice_line_id) tracks_sold, cast(count(us.invoice_line_id) AS FLOAT) / ( SELECT COUNT(*) from usa_sold ) percentage_sold FROM usa_sold us INNER JOIN track t ON t.track_id = us.track_id INNER JOIN genre g ON g.genre_id = t.genre_id WHERE g.name = 'Rock' GROUP BY 1 ORDER BY 3 DESC LIMIT 5; <jupyter_output>Done. <jupyter_text>Rock, Alternative & Punk, and Metal have the most tracks sold. Adding more artists along the lines of The Rolling Stones and Nirvana would be beneficial after adding the three albums from above.# Analyzing Employee Sales PerformanceWrite a query that finds the total dollar amount of sales assigned to each sales support agent within the company. Add any extra attributes for that employee that you find are relevant to the analysis.<jupyter_code>%%sql SELECT SUM(i.total) total_sales, c.support_rep_id FROM invoice i INNER JOIN customer c ON c.customer_id = i.customer_id GROUP BY 2<jupyter_output>Done. <jupyter_text>There is data on three sales agents.<jupyter_code>%%sql SELECT e.first_name ||' '|| e.last_name employee_name, e.title FROM employee e<jupyter_output>Done. <jupyter_text>Confirming the three sales agents.<jupyter_code>%%sql WITH total_sales_report AS ( SELECT SUM(i.total) total, c.support_rep_id FROM invoice i INNER JOIN customer c ON c.customer_id = i.customer_id GROUP BY 2 ) SELECT e.first_name ||' '|| e.last_name employee_name, e.hire_date, SUM(tsr.total) total_sales FROM total_sales_report tsr INNER JOIN employee e ON e.employee_id = tsr.support_rep_id GROUP BY 1;<jupyter_output>Done. <jupyter_text>Steve started six months later, so that's why his numbers are a bit behind.# Analyzing Sales by Country<jupyter_code>%%sql WITH country_or_other AS ( SELECT CASE WHEN ( SELECT count(*) FROM customer where country = c.country ) = 1 THEN "Other" ELSE c.country END AS country, c.customer_id, il.* FROM invoice_line il INNER JOIN invoice i ON i.invoice_id = il.invoice_id INNER JOIN customer c ON c.customer_id = i.customer_id ) SELECT country, customers, total_sales, average_order, customer_lifetime_value FROM ( SELECT country, count(distinct customer_id) customers, SUM(unit_price) total_sales, SUM(unit_price) / count(distinct customer_id) customer_lifetime_value, SUM(unit_price) / count(distinct invoice_id) average_order, CASE WHEN country = "Other" THEN 1 ELSE 0 END AS sort FROM country_or_other GROUP BY country ORDER BY sort ASC, total_sales DESC );<jupyter_output>Done.
no_license
/chinook_store_sql.ipynb
EdsTyping/chinook_record_store_sql
8
<jupyter_start><jupyter_text>## Surrogate Models & Helper Functions<jupyter_code>ValueRange = namedtuple('ValueRange', ['min', 'max']) def determinerange(values): """Determine the range of values in each dimension""" return ValueRange(np.min(values, axis=0), np.max(values, axis=0)) def linearscaletransform(values, *, range_in=None, range_out=ValueRange(0, 1), scale_only=False): """Perform a scale transformation of `values`: [range_in] --> [range_out]""" if range_in is None: range_in = determinerange(values) elif not isinstance(range_in, ValueRange): range_in = ValueRange(*range_in) if not isinstance(range_out, ValueRange): range_out = ValueRange(*range_out) scale_out = range_out.max - range_out.min scale_in = range_in.max - range_in.min if scale_only: scaled_values = (values / scale_in) * scale_out else: scaled_values = (values - range_in.min) / scale_in scaled_values = (scaled_values * scale_out) + range_out.min return scaled_values ''' F16 ''' def F16(X): f = bn.F16() X = np.array(X) return f(X) ''' Latin HyperCube Sampling Design of Experiment ''' def DOE(n_obs, dim): np.random.seed(0) lhd = pyDOE.lhs(n=dim, samples=n_obs, criterion='m') X = [lhd[:,idx] for idx in range(dim)] return X def create_basis_function(data): true = np.array(data['Y']) data = pd.DataFrame(np.atleast_2d(PolynomialFeatures(degree=2).fit_transform(data.iloc[:,:-1]))) data['Y'] = pd.Series(true) return data ''' Create Basis Functions ''' def create_function_basis(x): return np.atleast_2d(PolynomialFeatures(degree=2).fit_transform(x.reshape(1,-1))) ''' Elastic Net Regression ''' def elastic_net(train_data,test_data): scaler = MinMaxScaler().fit(np.r_[train_data.iloc[:,:-1].values]) regr = ElasticNet(alpha= 0.12 ,random_state=0 , l1_ratio=0.81, fit_intercept =True, max_iter=3000,selection='random').fit(scaler.transform ( np.array(train_data.iloc[:,:-1])) , np.array(train_data.iloc[:,-1])) pred = regr.predict(scaler.transform(test_data)) def predict(scaler, regr): def __predict__(x): x = create_function_basis(x) return regr.predict(scaler.transform(x)) return __predict__ return regr,pred, predict(scaler, regr) ''' Kriging''' def kriging(train_data,test_data): kernel = RBF() scaler = MinMaxScaler().fit(np.r_[train_data.iloc[:,:-1].values]) gpr = GaussianProcessRegressor(kernel=kernel,n_restarts_optimizer= 15,random_state=0, normalize_y=True ).fit(scaler.transform(train_data.iloc[:,:-1]), train_data.iloc[:,-1]) pred = gpr.predict(scaler.transform(test_data)) def predict(scaler, gpr): def __predict__(x): x = np.atleast_2d(x) return gpr.predict(scaler.transform(x)) return __predict__ return gpr,pred, predict(scaler,gpr) ''' Support Vector Regression''' def _SVR(train_data,test_data): scaler = MinMaxScaler().fit(np.r_[train_data.iloc[:,:-1].values]) gpr = sklearn.svm.SVR(kernel='rbf', gamma = 37.213462 , C = 1000.000000 ,max_iter=1500).fit( scaler.transform(train_data.iloc[:,:-1]), train_data.iloc[:,-1]) pred = gpr.predict(scaler.transform(test_data)) def predict(scaler, gpr): def __predict__(x): x = np.atleast_2d(x) return gpr.predict(scaler.transform(x)) return __predict__ return gpr,pred, predict(scaler,gpr)<jupyter_output><empty_output><jupyter_text>## Load Training and Test Data Set initially Generated<jupyter_code>path = "train_16_4000Samples.csv" train = pd.read_csv(path).iloc[:,1:] test = pd.read_csv('test_16_800Samples.csv').iloc[:,1:] true = np.array(test['Y'])<jupyter_output><empty_output><jupyter_text>## Surrogate Models<jupyter_code>model_kri , pred_kri , predict_kri = kriging(train,test.iloc[:,:-1]) model_svr , pred_svr , predict_svr = _SVR(train,test.iloc[:,:-1]) train = create_basis_function(train) test = create_basis_function(test) model_eln , pred_eln , predict_eln = elastic_net(train,test.iloc[:,:-1])<jupyter_output>/usr/local/lib/python3.6/dist-packages/sklearn/svm/_base.py:231: ConvergenceWarning: Solver terminated early (max_iter=1500). Consider pre-processing your data with StandardScaler or MinMaxScaler. % self.max_iter, ConvergenceWarning) <jupyter_text>## CMA-ES<jupyter_code>Columns = ['Kri' , 'SVR' , 'ELN' ] Cols = [] for j in range(len(Columns)): for i in range(1,201): Cols.append(Columns[j]+'_X'+str(i)) const = [ [-5] * 200, [5] * 200 ] opt = cma.CMAOptions() opt.set("bounds", const) opt.set ("seed" , 0) opt.set ("maxfevals" , 200000) n_obs , dim = 30, 200 G = DOE(n_obs, dim) G = [ linearscaletransform(G[idx] , range_out=(-5,5)) for idx in range(dim) ] G = [ G[idx].reshape(n_obs,1) for idx in range(len(G)) ] X_Values = np.zeros([10,600]) for i in range(10, 20): print ('Run : '+ str(i)) min_kri = cma.fmin(predict_kri , np.concatenate(G, 1)[i] , 2.5 , options=opt) [0] min_svr = cma.fmin(predict_svr , np.concatenate(G, 1)[i] , 2.5 , options=opt) [0] min_eln = cma.fmin(predict_eln , np.concatenate(G, 1)[i] , 2.5 , options=opt) [0] X_Values [i-10,:] = list(min_kri)+list(min_svr)+list(min_eln) X_Values = pd.DataFrame(X_Values) X_Values.columns = Cols X_Values.to_csv('sample_data\\F16_X_Values_first.csv') Krig_Fun = np.zeros(10) SVM_Fun = np.zeros(10) ELN_Fun = np.zeros(10) for i in range(X_Values.shape[0]): Krig_Fun [i] = F16(X_Values.iloc[i,:200]) SVM_Fun [i] = F16(X_Values.iloc[i,200:400]) ELN_Fun [i] = F16(X_Values.iloc[i,400:600]) print ('Kriging') print (stats.mode(Krig_Fun) , np.std(Krig_Fun)) print ('SVM') print (stats.mode(SVM_Fun) , np.std(SVM_Fun)) print ('ELN') print (stats.mode(ELN_Fun) , np.std(ELN_Fun)) <jupyter_output><empty_output>
no_license
/Original - Optimality/200D/second/F16_200_original.ipynb
SibghatUllah13/Deep-Latent_Variable_Models-for-dimensionality-reduction-in-surrogate-assisted-optimization
4
<jupyter_start><jupyter_text>## Download rnn_merged.zip & rnn_embed.zip from https://drive.google.com/drive/folders/1yO_W-m0fF_PludrnScdgyTGsPFoDsA6_?usp=sharing and unzip to the same folder of this file ## Also download train_jpg.zip & test_jpg.zip from competition website<jupyter_code>import pandas as pd import tensorflow as tf from keras.preprocessing import text, sequence import numpy as np from keras.layers import Input, SpatialDropout1D,Dropout, GlobalAveragePooling1D, GlobalMaxPooling1D, \ CuDNNGRU, GRU, Bidirectional, LSTM, Dense, Embedding, concatenate, Embedding, \ Flatten, Activation, BatchNormalization, regularizers, Conv1D, Conv2D, MaxPooling2D from keras.constraints import max_norm from keras.initializers import Orthogonal from keras.models import Model from keras.optimizers import Adam from keras.callbacks import ModelCheckpoint, LambdaCallback, Callback, LearningRateScheduler import keras.backend as K import numpy as np from sklearn import metrics from sklearn.model_selection import train_test_split import os import pickle import gc; gc.enable() import matplotlib import matplotlib.pyplot as plt %matplotlib inline import string import nltk from nltk.corpus import stopwords from nltk.stem.snowball import RussianStemmer from scipy.stats import boxcox import re #from tqdm import tqdm<jupyter_output>E:\Anaconda3\envs\tensorflow\lib\site-packages\h5py\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`. from ._conv import register_converters as _register_converters Using TensorFlow backend. <jupyter_text>### Check GPU Availability<jupyter_code>sess = tf.Session(config=tf.ConfigProto(log_device_placement=True)) from tensorflow.python.client import device_lib print(device_lib.list_local_devices()) K.tensorflow_backend._get_available_gpus()<jupyter_output>[name: "/device:CPU:0" device_type: "CPU" memory_limit: 268435456 locality { } incarnation: 339512226104842527 , name: "/device:GPU:0" device_type: "GPU" memory_limit: 3174131302 locality { bus_id: 1 links { } } incarnation: 17746058336949755705 physical_device_desc: "device: 0, name: GeForce GTX 1050 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1" ] <jupyter_text>### Preprocess Training and Testing Data<jupyter_code>seed = 411 rnn_train_epochs = 10 batch_size=128 # 32 or 64 is good (too huge for my PC), 128 is worse in the past experiments cpu_count=4 features = pickle.load(open('rnn_merged.pkl', 'rb')) features.keys() train = features['train'] test = features['test'] renamed_cols = [] count = 0 for col in train.columns: if 'cat_features_user_id_category_name' in col: col = 'cat_features_user_id_category_name_'+str(count) count += 1 renamed_cols.append(col) train.columns = renamed_cols test.columns = renamed_cols train_len = train.shape[0] train_y = features['y_train'] categorical = features['categorical'] numerical = [f for f in train.columns if f not in categorical] features = numerical + categorical train.columns.tolist() # remove features: text, image, other embeddings\feature engineerings remove_cols = [ 'cat_features_user_id_category_name_0', 'cat_features_user_id_category_name_1', 'cat_features_user_id_category_name_2', 'cat_features_user_id_category_name_3', 'cat_features_user_id_category_name_4', 'cat_features_user_id_category_name_5', 'cat_features_user_id_category_name_6', 'cat_features_user_id_category_name_7', 'cat_features_user_id_category_name_8', 'cat_features_user_id_category_name_9', 'cat_features_user_id_category_name_10', 'cat_features_user_id_category_name_11', 'cat_features_user_id_category_name_12', 'cat_features_user_id_category_name_13', 'cat_features_user_id_category_name_14', 'cat_features_user_id_category_name_15', 'cat_features_user_id_category_name_16', 'cat_features_user_id_category_name_17', 'cat_features_user_id_category_name_18', 'cat_features_user_id_category_name_19', 'cat_features_user_id_category_name_20', 'cat_features_user_id_category_name_21', 'cat_features_user_id_category_name_22', 'cat_features_user_id_category_name_23', 'cat_features_user_id_category_name_24', 'cat_features_user_id_category_name_25', 'cat_features_user_id_category_name_26', 'cat_features_user_id_category_name_27', 'cat_features_user_id_category_name_28', 'cat_features_user_id_category_name_29', 'cat_features_user_id_category_name_30', 'cat_features_user_id_category_name_31', 'cat_features_user_id_category_name_32', 'cat_features_user_id_category_name_33', 'cat_features_user_id_category_name_34', 'cat_features_user_id_category_name_35', 'cat_features_user_id_category_name_36', 'cat_features_user_id_category_name_37', 'cat_features_user_id_category_name_38', 'cat_features_user_id_category_name_39', 'cat_features_user_id_category_name_40', 'cat_features_user_id_category_name_41', 'cat_features_user_id_category_name_42', 'cat_features_user_id_category_name_43', 'cat_features_user_id_category_name_44', 'cat_features_user_id_category_name_45', 'cat_features_user_id_category_name_46', 'title_tfidf_svd_1', 'title_tfidf_svd_2', 'title_tfidf_svd_3', 'title_tfidf_svd_4', 'title_tfidf_svd_5', 'description_tfidf_svd_1', 'description_tfidf_svd_2', 'description_tfidf_svd_3', 'description_tfidf_svd_4', 'description_tfidf_svd_5', 'region_mean_price', 'region_mean_image_top_1', 'region_mean_item_seq_number', 'region_mean_price_pred', 'region_mean_price_pred_all', 'region_mean_ridge_preds', 'city_mean_price', 'city_mean_image_top_1', 'city_mean_item_seq_number', 'city_mean_price_pred', 'city_mean_price_pred_all', 'city_mean_ridge_preds', 'parent_category_name_mean_price', 'parent_category_name_mean_image_top_1', 'parent_category_name_mean_item_seq_number', 'parent_category_name_mean_price_pred', 'parent_category_name_mean_price_pred_all', 'parent_category_name_mean_ridge_preds', 'category_name_mean_price', 'category_name_mean_image_top_1', 'category_name_mean_item_seq_number', 'category_name_mean_price_pred', 'category_name_mean_price_pred_all', 'category_name_mean_ridge_preds', 'user_type_mean_price', 'user_type_mean_image_top_1', 'user_type_mean_item_seq_number', 'user_type_mean_price_pred', 'user_type_mean_price_pred_all', 'user_type_mean_ridge_preds', 'param_1_mean_price', 'param_1_mean_image_top_1', 'param_1_mean_item_seq_number', 'param_1_mean_price_pred', 'param_1_mean_price_pred_all', 'param_1_mean_ridge_preds', 'param_2_mean_price', 'param_2_mean_image_top_1', 'param_2_mean_item_seq_number', 'param_2_mean_price_pred', 'param_2_mean_price_pred_all', 'param_2_mean_ridge_preds', 'param_3_mean_price', 'param_3_mean_image_top_1', 'param_3_mean_item_seq_number', 'param_3_mean_price_pred', 'param_3_mean_price_pred_all', 'param_3_mean_ridge_preds', 'user_id_nunique_parent_category_name', 'user_id_nunique_category_name', 'user_id_nunique_param_1', 'user_id_nunique_param_2', 'user_id_nunique_param_3', 'user_id_nunique_activation_date', 'user_id_activation_date_count_item_id', 'image_top_1_nunique_item_id', 'image_top_1_nunique_user_id', 'image_top_1_nunique_category_name', 'image_top_1_nunique_param_1', 'image_top_1_nunique_item_seq_number', 'image_top_1_mean_price_pred', 'image_top_1_std_price_pred', 'image_top_1_mean_item_seq_number', 'user_id_mean_ridge_preds', 'user_id_category_name_mean_ridge_preds', 'user_id_image_top_1_mean_ridge_preds', 'user_id_category_name_sum_ridge_preds', 'cityxcatxusertypeitem_num', 'cityxcatxusertypecity_fm_factor_0', 'cityxcatxusertypecity_fm_factor_1', 'cityxcatxusertypecategory_name_fm_factor_0', 'cityxcatxusertypecategory_name_fm_factor_1', 'cityxcatxusertypeuser_type_fm_factor_0', 'cityxcatxusertypeuser_type_fm_factor_1', 'cityxcatxusertypecity_fm_bias', 'cityxcatxusertypecategory_name_fm_bias', 'cityxcatxusertypeuser_type_fm_bias', 'imgxcityxcatitem_num', 'imgxcityxcatimage_top_1_fm_factor_0', 'imgxcityxcatimage_top_1_fm_factor_1', 'imgxcityxcatcity_fm_factor_0', 'imgxcityxcatcity_fm_factor_1', 'imgxcityxcatcategory_name_fm_factor_0', 'imgxcityxcatcategory_name_fm_factor_1', 'imgxcityxcatimage_top_1_fm_bias', 'imgxcityxcatcity_fm_bias', 'imgxcityxcatcategory_name_fm_bias', 'imgxisqnxusertypeitem_num', 'imgxisqnxusertypeimage_top_1_fm_factor_0', 'imgxisqnxusertypeimage_top_1_fm_factor_1', 'imgxisqnxusertypeitem_seq_number_fm_factor_0', 'imgxisqnxusertypeitem_seq_number_fm_factor_1', 'imgxisqnxusertypeuser_type_fm_factor_0', 'imgxisqnxusertypeimage_top_1_fm_bias', 'imgxisqnxusertypeitem_seq_number_fm_bias', 'b_intensity_mean', 'b_intensity_median', 'b_intensity_std', 'g_intensity_mean', 'g_intensity_median', 'g_intensity_std', 'gray_intensity_mean', 'gray_intensity_median', 'gray_intensity_std', 'r_intensity_mean', 'r_intensity_median', 'r_intensity_std', 'nasnet_nima_med', 'nasnet_nima_std', 'nasnet_nima_max', 'nasnet_nima_min', 'nasnet_nima_1_quartile', 'nasnet_nima_3_quartile', 'nasnet_nima_13_quartile_diff', 'nasnet_nima_max_min_diff', 'nasnet_nima_non_max_mean', 'nasnet_nima_max_non_max_mean_diff', ] train.drop(remove_cols, axis=1, inplace=True) test.drop(remove_cols, axis=1, inplace=True) for col in remove_cols: if col in categorical: categorical.remove(col) if col in numerical: numerical.remove(col) features = numerical + categorical train.loc[:, 'image'] = pd.read_csv('train.csv', usecols=['activation_date', 'image'], parse_dates=['activation_date']) \ .sort_values('activation_date').reset_index(drop=True)['image'].fillna('no-image') test.loc[:, 'image'] = pd.read_csv('test.csv', usecols=['image'])['image'].fillna('no-image') max_features = 500000 maxlen = 150 embed_size = 300 title_max_features = 200000 title_maxlen = 80 title_embed_size = 100 embed_info = pickle.load(open('rnn_embed.pkl', 'rb')) embed_info.keys() desc_embed_info = embed_info['desc_embed_info'] title_embed_info = embed_info['title_embed_info'] print('setup max info for embedding in categorical variables') max_info = dict((col, train[col].max()+1) for col in categorical)<jupyter_output>setup max info for embedding in categorical variables <jupyter_text>### Build RNN Model<jupyter_code>def root_mean_squared_error(y_true, y_pred): return K.sqrt(K.mean(K.square(y_true - y_pred))) from keras.engine.topology import Layer from keras import initializers, regularizers, constraints class Attention(Layer): def __init__(self, step_dim, W_regularizer=None, b_regularizer=None, W_constraint=None, b_constraint=None, bias=True, **kwargs): self.supports_masking = True self.init = initializers.get('glorot_uniform') self.W_regularizer = regularizers.get(W_regularizer) self.b_regularizer = regularizers.get(b_regularizer) self.W_constraint = constraints.get(W_constraint) self.b_constraint = constraints.get(b_constraint) self.bias = bias self.step_dim = step_dim self.features_dim = 0 super(Attention, self).__init__(**kwargs) def build(self, input_shape): print(input_shape) assert len(input_shape) == 3 self.W = self.add_weight((input_shape[-1],), initializer=self.init, name='{}_W'.format(self.name), regularizer=self.W_regularizer, constraint=self.W_constraint) self.features_dim = input_shape[-1] if self.bias: self.b = self.add_weight((input_shape[1],), initializer='zero', name='{}_b'.format(self.name), regularizer=self.b_regularizer, constraint=self.b_constraint) else: self.b = None self.built = True def compute_mask(self, input, input_mask=None): return None def call(self, x, mask=None): features_dim = self.features_dim step_dim = self.step_dim eij = K.reshape(K.dot(K.reshape(x, (-1, features_dim)), K.reshape(self.W, (features_dim, 1))), (-1, step_dim)) if self.bias: eij += self.b eij = K.tanh(eij) a = K.exp(eij) if mask is not None: a *= K.cast(mask, K.floatx()) a /= K.cast(K.sum(a, axis=1, keepdims=True) + K.epsilon(), K.floatx()) a = K.expand_dims(a) weighted_input = x * a return K.sum(weighted_input, axis=1) def compute_output_shape(self, input_shape): return input_shape[0], self.features_dim def clip_rmse(true, prediction): return np.sqrt(metrics.mean_squared_error(true, np.clip(prediction, 0., 1.))) class NBatchEvalLogger(Callback): def __init__(self, display, val_X, val_y, save_path=None, save_start=1000): self.step = 0 self.display = display self.val_X = val_X self.val_y = val_y self.best_loss = None self.save_path = save_path self.save_start = save_start self.record_count = 0 def on_batch_end(self, batch, logs={}): self.step += 1 if self.step % self.display == 0 and self.step >= self.save_start: #loss, metric = self.model.evaluate(self.val_X, self.val_y, batch_size=128, verbose=1) prediction = self.model.predict(self.val_X, batch_size=128, verbose=0) loss = clip_rmse(self.val_y, prediction) if self.best_loss is None: self.best_loss = loss else: if loss < self.best_loss: self.best_loss = loss if self.save_path is not None: self.model.save(self.save_path, overwrite=True) self.record_count += 1 print('\rstep: {} val loss={:.5f}, best loss={:.5f}'.format(self.step, loss, self.best_loss)) import keras from copy import deepcopy as cp import os from zipfile import ZipFile import cv2 import numpy as np import pandas as pd from dask import bag, threaded from dask.diagnostics import ProgressBar import matplotlib.pyplot as plt from keras.preprocessing.image import load_img, img_to_array from keras.applications.resnet50 import preprocess_input import concurrent.futures from multiprocessing.pool import ThreadPool class DataGenerator(keras.utils.Sequence): #'Generates data for Keras' def __init__(self, list_IDs, X, y, img_arch, img_path, batch_size=32, shuffle=True, is_train=True): #'Initialization' self.batch_size = batch_size self.X = X self.y = y self.list_IDs = list_IDs self.shuffle = shuffle self.img_path = img_path self.is_train = is_train self.on_epoch_end() self.zipped = ZipFile(img_arch) #print('file names:\n', self.zipped.namelist()[1:10], '\n...') self.img_path = img_path global cpu_count self.pool = ThreadPool(cpu_count) def __getstate__(self): """ This is called before pickling. """ state = self.__dict__.copy() del state['zipped'] return state def __setstate__(self, state): """ This is called while unpickling. """ self.__dict__.update(state) def __len__(self): #'Denotes the number of batches per epoch' return int(np.ceil(len(self.list_IDs) / self.batch_size)) def __getitem__(self, index): #'Generate one batch of data' # Generate indexes of the batch start = index*self.batch_size end = min((index+1)*self.batch_size, len(self.indexes)) indexes = self.indexes[start: end] # Generate data return self.__data_generation(indexes) def on_epoch_end(self): #'Updates indexes after each epoch' self.indexes = cp(list(self.list_IDs)) if self.shuffle == True: np.random.shuffle(self.indexes) def load_img_from_zipped(self, img_id, i, imgs_holder): invalid_img_ids = ['4f029e2a00e892aa2cac27d98b52ef8b13d91471f613c8d3c38e3f29d4da0b0c', '8513a91e55670c709069b5f85e12a59095b802877715903abef16b7a6f306e58', '60d310a42e87cdf799afcd89dc1b11ae3fdc3d0233747ec7ef78d82c87002e83', 'b98b291bd04c3d92165ca515e00468fd9756af9a8f1df42505deed1dcfb5d7ae'] try: if img_id in invalid_img_ids or img_id == 'no-image': pass else: exfile = self.zipped.read(self.img_path+img_id+'.jpg') arr = np.frombuffer(exfile, np.uint8) imz = cv2.imdecode(arr, flags=cv2.IMREAD_UNCHANGED) imz = cv2.resize(imz, (224,224), interpolation=cv2.INTER_AREA) imgs_holder[i] = img_to_array(imz) except: print(img_id, ' is invalid') pass return None def parallel_load_imgs(self, img_ids, wait=True): imgs_holder = np.zeros((len(img_ids), 224, 224, 3)) ''' for i, im_id in enumerate(img_ids): self.load_img_from_zipped(im_id, i, imgs_holder) ''' self.res = [self.pool.apply_async(self.load_img_from_zipped, (im_id, i, imgs_holder)) for i, im_id in enumerate(img_ids)] if wait: for r in self.res: r.get() #print(imgs_holder) imgs_holder = preprocess_input(imgs_holder) # adjust to mean of rgb to some value return imgs_holder def __data_generation(self, list_IDs_temp): #'Generates data containing batch_size samples' # X : (n_samples, *dim, n_channels) # Generate data X = dict((col, self.X.loc[list_IDs_temp, col].values) for col in features) X['desc'] = desc_embed_info['text'][list_IDs_temp,:] X['title'] = title_embed_info['text'][list_IDs_temp,:] X['imgs'] = self.parallel_load_imgs(self.X.loc[list_IDs_temp, 'image'].values) if self.is_train: y = cp(self.y[list_IDs_temp]) return X, y else: return X # 'train_jpg.zip', 'data/competition_files/train_jpg/', # debug use ''' zipped = ZipFile('train_jpg.zip') print(zipped.namelist()[1:10]) img_id = '2809fd6afd6d3cae4dd4ad93a7f905a0db32292f4df4b3f19fa5492e08cbfd90' target_size=(224,224) try: exfile = zipped.read('data/competition_files/train_jpg/'+img_id+'.jpg') arr = np.frombuffer(exfile, np.uint8) imz = cv2.imdecode(arr, flags=cv2.IMREAD_UNCHANGED) imz = cv2.resize(imz, target_size, interpolation=cv2.INTER_AREA) except: print(img_id, ' is invalid') imz = None imz ''' def build_model(categorical_features, numerical_features): # non-cat features non_cat_inputs = [] for col in numerical_features: f = Input(shape=[1], name=col) non_cat_inputs.append(f) # cat features cat_inputs = [] cat_embeds = [] for col in categorical_features: f = Input(shape=[1], name=col) embed_dim = max_info[col].max() if max_info[col] > 10: reduced_dim = 10 else: reduced_dim = 1 embed_f = Embedding(embed_dim, reduced_dim)(f) flatten_f = Flatten()(embed_f) cat_inputs.append(f) cat_embeds.append(flatten_f) # text features: architecture of text to try here!!! # description text_inp = Input(shape = (maxlen, ), name='desc') text_emb = Embedding(desc_embed_info['nb_words'], embed_size, weights = [desc_embed_info['emb_matrix']], input_length = maxlen, trainable = False)(text_inp) text_emb = SpatialDropout1D(0.3)(text_emb) text_gru = Bidirectional(CuDNNGRU(128, return_sequences = True))(text_emb) text_gru = Conv1D(64, kernel_size = 3, padding = "valid", kernel_initializer = "glorot_uniform")(text_gru) text_gru_avg = GlobalAveragePooling1D()(text_gru) text_gru_max = GlobalMaxPooling1D()(text_gru) text_gru = concatenate([text_gru_avg, text_gru_max]) text_gru = Dropout(0.1)(text_gru) # title title_inp = Input(shape = (title_maxlen, ), name='title') title_emb = Embedding(title_embed_info['nb_words'], title_embed_size, weights = [title_embed_info['emb_matrix']], input_length = title_maxlen, trainable = False)(title_inp) title_emb = SpatialDropout1D(0.1)(title_emb) title_gru = Bidirectional(CuDNNGRU(32, return_sequences = True))(title_emb) title_gru = Conv1D(16, kernel_size = 3, padding = "valid", kernel_initializer = "glorot_uniform")(title_gru) title_gru_avg = GlobalAveragePooling1D()(title_gru) title_gru_max = GlobalMaxPooling1D()(title_gru) title_gru = concatenate([title_gru_avg, title_gru_max]) title_gru = Dropout(0.1)(title_gru) # add image architecture # reference: https://keras.io/getting-started/functional-api-guide/#more-examples, Visual question answering model ''' img_inp = Input(shape = (224, 224, 3 ), name='imgs') img_ch = Conv2D(64, (3, 3), activation='relu', padding='same', W_constraint=max_norm(3))(img_inp) img_ch = Conv2D(64, (3, 3), activation='relu')(img_ch) img_ch = MaxPooling2D((2, 2))(img_ch) #img_ch = Conv2D(128, (3, 3), activation='relu', padding='same', W_constraint=max_norm(3))(img_ch) #img_ch = Conv2D(128, (3, 3), activation='relu')(img_ch) #img_ch = MaxPooling2D((2, 2))(img_ch) #img_ch = Conv2D(256, (3, 3), activation='relu', padding='same', W_constraint=max_norm(3))(img_ch) #img_ch = Conv2D(256, (3, 3), activation='relu')(img_ch) #img_ch = Conv2D(256, (3, 3), activation='relu')(img_ch) #img_ch = MaxPooling2D((2, 2))(img_ch) img_ch = Flatten()(img_ch) img_ch = Dense(64, activation='relu')(img_ch) ''' # merge each branch: non-cat, cat, text, img concat_main = non_cat_inputs+cat_embeds+[text_gru, title_gru] main = concatenate(concat_main) main = BatchNormalization()(main) main = Dropout(0.1)(main) main = BatchNormalization()(Dense(256, activation='relu')(main)) main = Dropout(0.1)(main) main = BatchNormalization()(Dense(64, activation='relu')(main)) out = Dense(1, activation = "sigmoid")(main) concat_input = non_cat_inputs+cat_inputs+[text_inp, title_inp] model = Model(concat_input, out) model.regularizers = [regularizers.l2(0.0001)] model.compile(optimizer = Adam(lr=0.001), loss = root_mean_squared_error, metrics =[root_mean_squared_error]) model.summary() return model<jupyter_output><empty_output><jupyter_text>### Training<jupyter_code>from sklearn.model_selection import KFold import warnings; warnings.filterwarnings('ignore') train_indices = np.arange(0, train_len) test_indices = np.arange(0, test.shape[0]) from keras_tqdm import TQDMNotebookCallback from ipywidgets import IntProgress start_fold = 0 # <= 0 for invalid, train from fold 1, > 0: used to train from fold=start_fold resume_file_prefix = '0619_rnn' # whatever we like if start_fold > 0: import pickle ret = pickle.load(open(resume_file_prefix+'_oof_val_pred', 'rb')) ret_test = pickle.load(open(resume_file_prefix+'_oof_test_pred', 'rb')) print(ret) print(ret_test) else: ret = np.zeros((train.shape[0],)) ret_test = np.zeros((test.shape[0],)) fold = 0 for tr_ix, val_ix in KFold(5, shuffle=True, random_state=seed).split(train_indices): fold += 1 if start_fold > 0 and fold < start_fold: continue else: pass model = build_model(categorical, numerical) file_path = "rnn_weights/model_final_fold_{}.hdf5".format(fold) # customized batch loader training_generator = DataGenerator(tr_ix, train, train_y, 'train_jpg.zip', 'data/competition_files/train_jpg/', batch_size=batch_size, shuffle=True) validation_generator = DataGenerator(val_ix, train, train_y, 'train_jpg.zip', 'data/competition_files/train_jpg/', batch_size=batch_size, shuffle=False) lr_schd = LearningRateScheduler(lambda epoch: 0.001*(0.2**(epoch//6)), verbose=1) check_point = ModelCheckpoint(file_path, monitor = "val_loss", mode = "min", save_best_only = True, verbose = 1) history = model.fit_generator(generator=training_generator, validation_data=validation_generator, use_multiprocessing=False, workers=1, epochs=rnn_train_epochs, verbose = 0, callbacks = [lr_schd, check_point, TQDMNotebookCallback(leave_inner=True, leave_outer=True)]) # Predict val + test oofs model.load_weights(file_path) # load weight with best validation score del validation_generator validation_generator = DataGenerator(val_ix, train, None, 'train_jpg.zip', 'data/competition_files/train_jpg/', batch_size=batch_size, shuffle=False, is_train=False) test_generator = DataGenerator(test_indices, test, None, 'test_jpg.zip', 'data/competition_files/test_jpg/', batch_size=batch_size, shuffle=False, is_train=False) ret[val_ix] = model.predict_generator(validation_generator, use_multiprocessing=False, workers=1).reshape((len(val_ix),)) ret_test += model.predict_generator(test_generator, use_multiprocessing=False, workers=1).reshape((ret_test.shape[0],)) del model, history, training_generator, validation_generator, test_generator; gc.collect() ret_test /= 5 # uncomment these to dump files if OOM (out-of-mem) happens import pickle pickle.dump(ret, open(resume_file_prefix+'_oof_val_pred', 'wb')) pickle.dump(ret_test, open(resume_file_prefix+'_oof_test_pred', 'wb')) # public kernel: cv = .2220, lb = .2247 # bigru-conv1d: cv =.2185 , lb = .2235 # bigru-attention: cv =.2186 , lb = .2235 # 2gru: lb: .2239 # self-trained wordvec: cv .217232, lb: .2229 # +partial new features: cv .216326, lb: <jupyter_output><empty_output><jupyter_text>### Generate OOFs and Submissions<jupyter_code>prefix = 'selftrained_bigru_conv1d_merged' pd.DataFrame(data=ret, columns=[prefix+'_rnn_pred']).to_csv(prefix+'_rnn_oof_val_pred.csv', index=False) pd.DataFrame(data=ret_test, columns=[prefix+'_rnn_pred']).to_csv(prefix+'_rnn_oof_test_pred.csv', index=False) subm = pd.read_csv('sample_submission.csv') subm['deal_probability'] = np.clip(ret_test, 0, 1) subm.to_csv(prefix+'_rnn_submission.csv', index=False)<jupyter_output><empty_output>
no_license
/RNN Self-Trained WordVec + Image + Merge Features (with Fast Loading)-Copy1.ipynb
tnmichael309/kaggle-avito-demand-challenge
6
<jupyter_start><jupyter_text># Consume deployed webservice via REST Demonstrates the usage of a deployed model via plain REST. REST is language-agnostic, so you should be able to query from any REST-capable programming language.## Configuration<jupyter_code>from environs import Env env = Env() env.read_env("foundation.env") env.read_env("service-principals.env") # image to test IMAGE_TO_TEST = "mnist_fashion/04_consumption/random_test_images/random-test-image-1601.png" # endpoint of the scoring webservice SCORING_URI = "<...use your own..., eg. https://....westeurope.cloudapp.azure.com:443/api/v1/service/mnist-fashion-service/score>" # auth method, either "Token", "Keys" or "None". # also specify additional values depending on auth method AUTH_METHOD = "Token" if AUTH_METHOD == "Keys": AUTH_KEY = "<add your key here>" elif AUTH_METHOD == "Token": REGION = "eastus" SUBSCRIPTION_ID = env("SUBSCRIPTION_ID") RESOURCE_GROUP = env("RESOURCE_GROUP") WORKSPACE_NAME = env("WORKSPACE_NAME") SERVICE_NAME = "mnist-fashion-service" CONSUME_MODEL_SP_TENANT_ID = env("CONSUME_MODEL_SP_TENANT_ID") CONSUME_MODEL_SP_CLIENT_ID = env("CONSUME_MODEL_SP_CLIENT_ID") CONSUME_MODEL_SP_CLIENT_SECRET = env("CONSUME_MODEL_SP_CLIENT_SECRET") elif AUTH_METHOD == "None": pass<jupyter_output><empty_output><jupyter_text>## Load a random image and plot it<jupyter_code>import matplotlib.pyplot as plt from PIL import Image image = Image.open(IMAGE_TO_TEST) plt.figure() plt.imshow(image) plt.colorbar() plt.grid(False) plt.show()<jupyter_output><empty_output><jupyter_text>## Invoke the webservice and show result<jupyter_code>import requests import json # --- get input data input_data = open(IMAGE_TO_TEST, "rb").read() # alternatively for JSON input #input_data = json.dumps({"x": 4711}) # --- get headers # Content-Type # for binary data headers = {"Content-Type": "application/octet-stream"} # alternatively for JSON data #headers = {"Content-Type": "application/json"} # Authorization if AUTH_METHOD == "Token": # get an access token for the service principal to access Azure azure_access_token = requests.post( f"https://login.microsoftonline.com/{CONSUME_MODEL_SP_TENANT_ID}/oauth2/token", headers={"Content-Type": "application/x-www-form-urlencoded"}, data="grant_type=client_credentials" + "&resource=https%3A%2F%2Fmanagement.azure.com%2F" + f"&client_id={CONSUME_MODEL_SP_CLIENT_ID}" + f"&client_secret={CONSUME_MODEL_SP_CLIENT_SECRET}", ).json()["access_token"] # use that token to get another token for accessing the webservice # note: the token is only valid for a certain period of time. # after that time, a new token has to be used. the logic # to do this, is not implemented here yet. you can check # the current time against the refresh after time to know # if a new token is required. refreshAfter and expiryOn # are UNIX timestamps. use time.time() to get the current # timestamp. token_response = requests.post( f"https://{REGION}.modelmanagement.azureml.net/modelmanagement/v1.0/subscriptions/{SUBSCRIPTION_ID}/resourceGroups/{RESOURCE_GROUP}/providers/Microsoft.MachineLearningServices/workspaces/{WORKSPACE_NAME}/services/{SERVICE_NAME}/token", headers={"Authorization": f"Bearer {azure_access_token}"} ).json() access_token = token_response["accessToken"] access_token_refresh_after = int(token_response["refreshAfter"]) access_token_expiry_on = int(token_response["expiryOn"]) # finally, use that token to access the webservice headers["Authorization"] = f"Bearer {access_token}" if AUTH_METHOD == "Keys": headers["Authorization"] = f"Bearer {AUTH_KEY}" if AUTH_METHOD == "None": # do nothing pass # --- make request and display response response = requests.post(SCORING_URI, input_data, headers=headers, verify=True) print(response.json())<jupyter_output>{'predicted_label': 'Bag', 'confidence': '1.0'}
permissive
/mnist_fashion/04_consumption/consume-webservice.ipynb
anderl80/aml-template
3
<jupyter_start><jupyter_text># Problem : Print all ancestors of binary tree. Algorithm: 1. Check if root or node is None, if yes, return False 2. Append the ancestor list with the root 3. If the root equals node return True to the calling function 4. Check the left and right subtree for node recursively 5. If found return true else pop the node from the list and return False.\ Time Complexity: O(n) Space Complexity: O(n^2)<jupyter_code>def findAllAncestors(root, node, ancestors): if root is None or node is None: return False ancestors.append(root.getData()) if root.getData() == node: return True if (findAllAncestors(root.getLeft(), node, ancestors) or findAllAncestors(root.getRight(), node, ancestors)): return True ancestors.pop() return False class TreeNode: def __init__(self, data, left=None, right=None): self.data = data self.left = left self.right = right def getData(self): return self.data def setData(self, data): self.data = data def getLeft(self): return self.left def getRight(self): return self.right def setLeft(self, left): self.left = left def setRight(self, right): self.right = right a = TreeNode(1) b = TreeNode(2) c = TreeNode(3) a.setLeft(b) a.setRight(c) d = TreeNode(4) e = TreeNode(5) x = TreeNode(8) y = TreeNode(9) e.setLeft(x) b.setLeft(d) b.setRight(e) f = TreeNode(6) g = TreeNode(7) f.setRight(y) c.setLeft(f) c.setRight(g) ans = [] findAllAncestors(a, 9, ans) print(ans)<jupyter_output>[1, 3, 6, 9] <jupyter_text># Alternative way of solving the problem<jupyter_code>def printAllAncestors(root, node): if root is None or node is None: return False if ((root.getData() == node) or printAllAncestors(root.getLeft(), node) or printAllAncestors(root.getRight(), node)): print(root.getData()) return True return False printAllAncestors(a,9)<jupyter_output>9 6 3 1 <jupyter_text># Solution without recursion. Use postorder traversal to solve this problem. <jupyter_code>def findAncestorsIteratively(root, target): if root is None or target is None: return False stack = [] node = root prev_right = None while node or stack: if node: stack.append(node) node = node.getLeft() else: element = stack[-1] if element.getData() == target: break if element.getRight() and prev_right!=element.getRight(): prev_right= node = element.getRight() else: stack.pop() for i in stack: print(i.getData()) findAncestorsIteratively(a,9)<jupyter_output>1 3 6 9
no_license
/Trees/Binary Trees/Problems/.ipynb_checkpoints/AncestorsOfBinaryTree-checkpoint.ipynb
sumeet13/Algorithms-and-Data-Structures
3
<jupyter_start><jupyter_text>Table of Contents 1&nbsp;&nbsp;HISTOGRAM2&nbsp;&nbsp;QQPLOT3&nbsp;&nbsp;AGGREGATION PLOT (part4)## HISTOGRAM<jupyter_code>library(MASS) # Create a histogram of counts with hist() hist(Cars93$Horsepower, main = "hist() plot") # Create a normalized histogram with truehist() hist(Cars93$Horsepower, main = "truehist() plot", freq = FALSE) # Add the density curve to the histogram lines(density(Cars93$Horsepower))<jupyter_output><empty_output><jupyter_text>## QQPLOT<jupyter_code># Load the car package to make qqPlot() available library(car) # Create index16, pointing to 16-week chicks index16 <- which(ChickWeight$Time == 16) # Get the 16-week chick weights weights <- ChickWeight[index16,"weight"] # Show the normal QQ-plot of the chick weights qqPlot(weights) hist(weights, freq = FALSE) # Show the normal QQ-plot of the Boston$tax data qqPlot(Boston$tax) hist(Boston$tax)<jupyter_output>Warning message: "package 'car' was built under R version 3.5.2"<jupyter_text>## AGGREGATION PLOT (part4)<jupyter_code># Set up a two-by-two plot array par(mfrow = c(2,2)) # Plot the raw duration data plot(geyser$duration, main = "Raw data") # Plot the normalized histogram of the duration data truehist(geyser$duration, main = "Histogram") # Plot the density of the duration data plot(density(geyser$duration), main = "Density") # Construct the normal QQ-plot of the duration data qqPlot(geyser$duration, main = "QQ-plot")<jupyter_output><empty_output>
no_license
/Data visualization - base R/.ipynb_checkpoints/2.1. One variable-checkpoint.ipynb
yoogun143/Datacamp_R
3
<jupyter_start><jupyter_text>原文代码作者:François Chollet github:https://github.com/fchollet/deep-learning-with-python-notebooks 中文注释制作:黄海广 github:https://github.com/fengdu78 代码全部测试通过。 配置环境:keras 2.2.1(原文是2.0.8,运行结果一致),tensorflow 1.8,python 3.6, 主机:显卡:一块1080ti;内存:32g(注:绝大部分代码不需要GPU) ![公众号](data/gongzhong.jpg)<jupyter_code>import keras keras.__version__<jupyter_output>Using TensorFlow backend. <jupyter_text># Using word embeddings # 使用词嵌入 This notebook contains the second code sample found in Chapter 6, Section 1 of [Deep Learning with Python](https://www.manning.com/books/deep-learning-with-python?a_aid=keras&a_bid=76564dff). Note that the original text features far more content, in particular further explanations and figures: in this notebook, you will only find source code and related comments. --- Another popular and powerful way to associate a vector with a word is the use of dense "word vectors", also called "word embeddings". While the vectors obtained through one-hot encoding are binary, sparse (mostly made of zeros) and very high-dimensional (same dimensionality as the number of words in the vocabulary), "word embeddings" are low-dimensional floating point vectors (i.e. "dense" vectors, as opposed to sparse vectors). Unlike word vectors obtained via one-hot encoding, word embeddings are learned from data. It is common to see word embeddings that are 256-dimensional, 512-dimensional, or 1024-dimensional when dealing with very large vocabularies. On the other hand, one-hot encoding words generally leads to vectors that are 20,000-dimensional or higher (capturing a vocabulary of 20,000 token in this case). So, word embeddings pack more information into far fewer dimensions. 将单词与向量相关联还有另一种常用的强大方法,就是使用密集的词向量(word vector),也叫词嵌入(word embedding)。one-hot 编码得到的向量是二进制的、稀疏的(绝大部分元素都是 0)、维度很高的(维度大小等于词表中的单词个数),而词嵌入是低维的浮点数向量(即密集向量,与稀疏向量相对),参见下图。与 one-hot 编码得到的词向量不同,词嵌入是从数据中学习得到的。常见的词向量维度是 256、512 或 1024(处理非常大的词表时)。与此相对,one-hot 编码的词向量维度通常为 20 000 或更高(对应包含 20 000 个标记的词表)。因此,词向量可以将更多的信息塞入更低的维度中。 ![word embeddings vs. one hot encoding](data/word_embeddings.png)There are two ways to obtain word embeddings: * Learn word embeddings jointly with the main task you care about (e.g. document classification or sentiment prediction). In this setup, you would start with random word vectors, then learn your word vectors in the same way that you learn the weights of a neural network. * Load into your model word embeddings that were pre-computed using a different machine learning task than the one you are trying to solve. These are called "pre-trained word embeddings". Let's take a look at both. 获取词嵌入有两种方法。 * 在完成主任务(比如文档分类或情感预测)的同时学习词嵌入。在这种情况下,一开始是随机的词向量,然后对这些词向量进行学习,其学习方式与学习神经网络的权重相同。 * 在不同于待解决问题的机器学习任务上预计算好词嵌入,然后将其加载到模型中。这些词嵌入叫作预训练词嵌入(pretrained word embedding)。 我们来分别看一下这两种方法。 ## Learning word embeddings with the `Embedding` layer The simplest way to associate a dense vector to a word would be to pick the vector at random. The problem with this approach is that the resulting embedding space would have no structure: for instance, the words "accurate" and "exact" may end up with completely different embeddings, even though they are interchangeable in most sentences. It would be very difficult for a deep neural network to make sense of such a noisy, unstructured embedding space. To get a bit more abstract: the geometric relationships between word vectors should reflect the semantic relationships between these words. Word embeddings are meant to map human language into a geometric space. For instance, in a reasonable embedding space, we would expect synonyms to be embedded into similar word vectors, and in general we would expect the geometric distance (e.g. L2 distance) between any two word vectors to relate to the semantic distance of the associated words (words meaning very different things would be embedded to points far away from each other, while related words would be closer). Even beyond mere distance, we may want specific __directions__ in the embedding space to be meaningful. [...] In real-world word embedding spaces, common examples of meaningful geometric transformations are "gender vectors" and "plural vector". For instance, by adding a "female vector" to the vector "king", one obtain the vector "queen". By adding a "plural vector", one obtain "kings". Word embedding spaces typically feature thousands of such interpretable and potentially useful vectors. Is there some "ideal" word embedding space that would perfectly map human language and could be used for any natural language processing task? Possibly, but in any case, we have yet to compute anything of the sort. Also, there isn't such a thing as "human language", there are many different languages and they are not isomorphic, as a language is the reflection of a specific culture and a specific context. But more pragmatically, what makes a good word embedding space depends heavily on your task: the perfect word embedding space for an English-language movie review sentiment analysis model may look very different from the perfect embedding space for an English-language legal document classification model, because the importance of certain semantic relationships varies from task to task. It is thus reasonable to __learn__ a new embedding space with every new task. Thankfully, backpropagation makes this really easy, and Keras makes it even easier. It's just about learning the weights of a layer: the `Embedding` layer. ## 利用 Embedding 层学习词嵌入 要将一个词与一个密集向量相关联,最简单的方法就是随机选择向量。这种方法的问题在于,得到的嵌入空间没有任何结构。例如,accurate 和 exact 两个词的嵌入可能完全不同,尽管它们在大多数句子里都是可以互换的 a。深度神经网络很难对这种杂乱的、非结构化的嵌入空间进行学习。 说得更抽象一点,词向量之间的几何关系应该表示这些词之间的语义关系。词嵌入的作用应该是将人类的语言映射到几何空间中。例如,在一个合理的嵌入空间中,同义词应该被嵌入到相似的词向量中,一般来说,任意两个词向量之间的几何距离(比如 L2 距离)应该和这两个词的语义距离有关(表示不同事物的词被嵌入到相隔很远的点,而相关的词则更加靠近)。除了距离,你可能还希望嵌入空间中的特定方向也是有意义的。 在真实的词嵌入空间中,常见的有意义的几何变换的例子包括“性别”向量和“复数”向量。 例如,将 king(国王)向量加上 female(女性)向量,得到的是 queen(女王)向量。将 king(国王) 向量加上 plural(复数)向量,得到的是 kings 向量。词嵌入空间通常具有几千个这种可解释的、 并且可能很有用的向量。 [...] 有没有一个理想的词嵌入空间,可以完美地映射人类语言,并可用于所有自然语言处理任 务?可能有,但我们尚未发现。此外,也不存在人类语言(human language)这种东西。世界上有许多种不同的语言,而且它们不是同构的,因为语言是特定文化和特定环境的反射。但从更实际的角度来说,一个好的词嵌入空间在很大程度上取决于你的任务。英语电影评论情感分析 模型的完美词嵌入空间,可能不同于英语法律文档分类模型的完美词嵌入空间,因为某些语义关系的重要性因任务而异。 因此,合理的做法是对每个新任务都学习一个新的嵌入空间。幸运的是,反向传播让这种学习变得很简单,而 Keras 使其变得更简单。我们要做的就是学习一个层的权重,这个层就是 Embedding 层。 <jupyter_code>from keras.layers import Embedding # The Embedding layer takes at least two arguments: # the number of possible tokens, here 1000 (1 + maximum word index), # and the dimensionality of the embeddings, here 64. # Embedding 层至少需要两个参数: # 标记的个数(这里是 1000,即最大单词索引 +1)和嵌入的维度(这里是 64) embedding_layer = Embedding(1000, 64)<jupyter_output><empty_output><jupyter_text> The `Embedding` layer is best understood as a dictionary mapping integer indices (which stand for specific words) to dense vectors. It takes as input integers, it looks up these integers into an internal dictionary, and it returns the associated vectors. It's effectively a dictionary lookup. 最好将 Embedding 层理解为一个字典,将整数索引(表示特定单词)映射为密集向量。它接收整数作为输入,并在内部字典中查找这些整数,然后返回相关联的向量。Embedding 层实际上是一种字典查找。 The `Embedding` layer takes as input a 2D tensor of integers, of shape `(samples, sequence_length)`, where each entry is a sequence of integers. It can embed sequences of variable lengths, so for instance we could feed into our embedding layer above batches that could have shapes `(32, 10)` (batch of 32 sequences of length 10) or `(64, 15)` (batch of 64 sequences of length 15). All sequences in a batch must have the same length, though (since we need to pack them into a single tensor), so sequences that are shorter than others should be padded with zeros, and sequences that are longer should be truncated. This layer returns a 3D floating point tensor, of shape `(samples, sequence_length, embedding_dimensionality)`. Such a 3D tensor can then be processed by a RNN layer or a 1D convolution layer (both will be introduced in the next sections). When you instantiate an `Embedding` layer, its weights (its internal dictionary of token vectors) are initially random, just like with any other layer. During training, these word vectors will be gradually adjusted via backpropagation, structuring the space into something that the downstream model can exploit. Once fully trained, your embedding space will show a lot of structure -- a kind of structure specialized for the specific problem you were training your model for. Let's apply this idea to the IMDB movie review sentiment prediction task that you are already familiar with. Let's quickly prepare the data. We will restrict the movie reviews to the top 10,000 most common words (like we did the first time we worked with this dataset), and cut the reviews after only 20 words. Our network will simply learn 8-dimensional embeddings for each of the 10,000 words, turn the input integer sequences (2D integer tensor) into embedded sequences (3D float tensor), flatten the tensor to 2D, and train a single `Dense` layer on top for classification. Embedding 层的输入是一个二维整数张量,其形状为 (samples, sequence_length),每个元素是一个整数序列。它能够嵌入长度可变的序列,例如,对于前一个例子中的Embedding 层,你可以输入形状为 (32, 10)(32 个长度为 10 的序列组成的批量)或 (64,15)(64 个长度为 15 的序列组成的批量)的批量。不过一批数据中的所有序列必须具有相同的 长度(因为需要将它们打包成一个张量),所以较短的序列应该用 0 填充,较长的序列应该被截断。 这个Embedding 层返回一个形状为 (samples, sequence_length, embedding_dimensionality) 的三维浮点数张量。然后可以用 RNN 层或一维卷积层来处理这个三维张量(二者都会在后面介绍)。 将一个Embedding 层实例化时,它的权重(即标记向量的内部字典)最开始是随机的,与其他层一样。在训练过程中,利用反向传播来逐渐调节这些词向量,改变空间结构以便下游模型可以利用。一旦训练完成,嵌入空间将会展示大量结构,这种结构专门针对训练模型所要解决的问题。 我们将这个想法应用于你熟悉的 IMDB 电影评论情感预测任务。首先,我们需要快速准备数据。将电影评论限制为前 10 000 个最常见的单词(第一次处理这个数据集时就是这么做的), 然后将评论长度限制为只有 20 个单词。对于这 10 000 个单词,网络将对每个词都学习一个 8 维嵌入,将输入的整数序列(二维整数张量)转换为嵌入序列(三维浮点数张量),然后将这个张量展平为二维,最后在上面训练一个Dense 层用于分类。 <jupyter_code>from keras.datasets import imdb from keras import preprocessing # Number of words to consider as features # 作为特征的单词个数 max_features = 10000 # Cut texts after this number of words # (among top max_features most common words) # 在这么多单词后截断文本(这些单词都属于前 max_features 个最常见的单词) maxlen = 20 # Load the data as lists of integers.(将数据加载为整数列表) (x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=max_features) # This turns our lists of integers # into a 2D integer tensor of shape `(samples, maxlen)` # 将整数列表转换成形状为 (samples, maxlen) 的二维整数张量 x_train = preprocessing.sequence.pad_sequences(x_train, maxlen=maxlen) x_test = preprocessing.sequence.pad_sequences(x_test, maxlen=maxlen) from keras.models import Sequential from keras.layers import Flatten, Dense model = Sequential() # We specify the maximum input length to our Embedding layer # so we can later flatten the embedded inputs # 指定 Embedding 层的最大输入长度,以便后面将嵌入输入展平。 model.add(Embedding(10000, 8, input_length=maxlen)) # After the Embedding layer, # our activations have shape `(samples, maxlen, 8)`. # Embedding 层 激活的形状为 (samples, maxlen, 8) # We flatten the 3D tensor of embeddings # into a 2D tensor of shape `(samples, maxlen * 8)` # 将三维的嵌入张量展平成形状为 (samples, maxlen * 8) 的二维张量 model.add(Flatten()) # We add the classifier on top model.add(Dense(1, activation='sigmoid')) model.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['acc']) model.summary() history = model.fit(x_train, y_train, epochs=10, batch_size=32, validation_split=0.2)<jupyter_output>_________________________________________________________________ Layer (type) Output Shape Param # ================================================================= embedding_2 (Embedding) (None, 20, 8) 80000 _________________________________________________________________ flatten_1 (Flatten) (None, 160) 0 _________________________________________________________________ dense_1 (Dense) (None, 1) 161 ================================================================= Total params: 80,161 Trainable params: 80,161 Non-trainable params: 0 _________________________________________________________________ Train on 20000 samples, validate on 5000 samples Epoch 1/10 20000/20000 [==============================] - 5s 237us/step - loss: 0.6759 - acc: 0.6050 - val_loss: 0.6398 - val_acc: 0.6814 Epoch 2/10 20000/20000 [==============================] - 3s 126us/step - loss: 0.5[...]<jupyter_text>We get to a validation accuracy of ~76%, which is pretty good considering that we only look at the first 20 words in every review. But note that merely flattening the embedded sequences and training a single `Dense` layer on top leads to a model that treats each word in the input sequence separately, without considering inter-word relationships and structure sentence (e.g. it would likely treat both _"this movie is shit"_ and _"this movie is the shit"_ as being negative "reviews"). It would be much better to add recurrent layers or 1D convolutional layers on top of the embedded sequences to learn features that take into account each sequence as a whole. That's what we will focus on in the next few sections. 得到的验证精度约为 76%,考虑到仅查看每条评论的前 20 个单词,这个结果还是相当不错的。但请注意,仅仅将嵌入序列展开并在上面训练一个 Dense 层,会导致模型对输入序列中的每个单词单独处理,而没有考虑单词之间的关系和句子结构(举个例子,这个模型可能会将 this movie is a bomb 和 this movie is the bomb 两条都归为负面评论)。更好的做法是在嵌入序列上添加循环层或一维卷积层,将每个序列作为整体来学习特征。这也是接下来几节的重点。 ## Using pre-trained word embeddings Sometimes, you have so little training data available that could never use your data alone to learn an appropriate task-specific embedding of your vocabulary. What to do then? Instead of learning word embeddings jointly with the problem you want to solve, you could be loading embedding vectors from a pre-computed embedding space known to be highly structured and to exhibit useful properties -- that captures generic aspects of language structure. The rationale behind using pre-trained word embeddings in natural language processing is very much the same as for using pre-trained convnets in image classification: we don't have enough data available to learn truly powerful features on our own, but we expect the features that we need to be fairly generic, i.e. common visual features or semantic features. In this case it makes sense to reuse features learned on a different problem. Such word embeddings are generally computed using word occurrence statistics (observations about what words co-occur in sentences or documents), using a variety of techniques, some involving neural networks, others not. The idea of a dense, low-dimensional embedding space for words, computed in an unsupervised way, was initially explored by Bengio et al. in the early 2000s, but it only started really taking off in research and industry applications after the release of one of the most famous and successful word embedding scheme: the Word2Vec algorithm, developed by Mikolov at Google in 2013. Word2Vec dimensions capture specific semantic properties, e.g. gender. There are various pre-computed databases of word embeddings that can download and start using in a Keras `Embedding` layer. Word2Vec is one of them. Another popular one is called "GloVe", developed by Stanford researchers in 2014. It stands for "Global Vectors for Word Representation", and it is an embedding technique based on factorizing a matrix of word co-occurrence statistics. Its developers have made available pre-computed embeddings for millions of English tokens, obtained from Wikipedia data or from Common Crawl data. Let's take a look at how you can get started using GloVe embeddings in a Keras model. The same method will of course be valid for Word2Vec embeddings or any other word embedding database that you can download. We will also use this example to refresh the text tokenization techniques we introduced a few paragraphs ago: we will start from raw text, and work our way up. ## 使用预训练的词嵌入 有时可用的训练数据很少,以至于只用手头数据无法学习适合特定任务的词嵌入。那么应该怎么办? 你可以从预计算的嵌入空间中加载嵌入向量(你知道这个嵌入空间是高度结构化的,并且具有有用的属性,即抓住了语言结构的一般特点),而不是在解决问题的同时学习词嵌入。在自然语言处理中使用预训练的词嵌入,其背后的原理与在图像分类中使用预训练的卷积神经网络是一样的:没有足够的数据来自己学习真正强大的特征,但你需要的特征应该是非常通用的,比如常见的视觉特征或语义特征。在这种情况下,重复使用在其他问题上学到的特征,这种做法是有道理的。 这种词嵌入通常是利用词频统计计算得出的(观察哪些词共同出现在句子或文档中),用到的技术很多,有些涉及神经网络,有些则不涉及。Bengio 等人在 21 世纪初首先研究了一种思路,就是用无监督的方法计算一个密集的低维词嵌入空间,但直到最有名且最成功的词嵌入方案之一 word2vec 算法发布之后,这一思路才开始在研究领域和工业应用中取得成功。word2vec 算法由 Google 的 Tomas Mikolov 于 2013 年开发,其维度抓住了特定的语义属性,比如性别。 有许多预计算的词嵌入数据库,你都可以下载并在 Keras 的 Embedding 层中使用。word2vec 就是其中之一。另一个常用的是 GloVe(global vectors for word representation,词表示全局向量),由斯坦福大学的研究人员于 2014 年开发。这种嵌入方法基于对词共现统计矩阵进行因式分解。其开发者已经公开了数百万个英文标记的预计算嵌入,它们都是从维基百科数据和 Common Crawl 数据得到的。 我们来看一下如何在 Keras 模型中使用 GloVe 嵌入。同样的方法也适用于 word2vec 嵌入或其他词嵌入数据库。这个例子还可以改进前面刚刚介绍过的文本分词技术,即从原始文本开始,一步步进行处理。 ## Putting it all together: from raw text to word embeddings We will be using a model similar to the one we just went over -- embedding sentences in sequences of vectors, flattening them and training a `Dense` layer on top. But we will do it using pre-trained word embeddings, and instead of using the pre-tokenized IMDB data packaged in Keras, we will start from scratch, by downloading the original text data. ## 整合在一起:从原始文本到词嵌入 本节的模型与之前刚刚见过的那个类似:将句子嵌入到向量序列中,然后将其展平,最后在上面训练一个 Dense 层。但此处将使用预训练的词嵌入。此外,我们将从头开始,先下载IMDB 原始文本数据,而不是使用 Keras 内置的已经预先分词的 IMDB 数据。 ### Download the IMDB data as raw text First, head to `http://ai.stanford.edu/~amaas/data/sentiment/` and download the raw IMDB dataset (if the URL isn't working anymore, just Google "IMDB dataset"). Uncompress it. Now let's collect the individual training reviews into a list of strings, one string per review, and let's also collect the review labels (positive / negative) into a `labels` list: ### 下载 IMDB 数据的原始文本 首先,打开 `http://ai.stanford.edu/~amaas/data/sentiment/` ,下载原始 IMDB 数据集并解压。 接下来,我们将训练评论转换成字符串列表,每个字符串对应一条评论。你也可以将评论 标签(正面 / 负面)转换成 labels 列表。 注:这一步我们已经做好了,不需要再下载,数据已经在 'data/aclImdb'目录。 <jupyter_code>import os imdb_dir = 'data/aclImdb' train_dir = os.path.join(imdb_dir, 'train') labels = [] texts = [] for label_type in ['neg', 'pos']: dir_name = os.path.join(train_dir, label_type) for fname in os.listdir(dir_name): if fname[-4:] == '.txt': f = open(os.path.join(dir_name, fname),encoding='utf8') texts.append(f.read()) f.close() if label_type == 'neg': labels.append(0) else: labels.append(1)<jupyter_output><empty_output><jupyter_text>### Tokenize the data Let's vectorize the texts we collected, and prepare a training and validation split. We will merely be using the concepts we introduced earlier in this section. Because pre-trained word embeddings are meant to be particularly useful on problems where little training data is available (otherwise, task-specific embeddings are likely to outperform them), we will add the following twist: we restrict the training data to its first 200 samples. So we will be learning to classify movie reviews after looking at just 200 examples... ### 对数据进行分词 利用本节前面介绍过的概念,我们对文本进行分词,并将其划分为训练集和验证集。 因为预训练的词嵌入对训练数据很少的问题特别有用(否则,针对于具体任务的嵌入可能效果更好),所以我们又添加了以下限制:将训练数据限定为前 200 个样本。因此,你需要在读取 200 个样本之后学习对电影评论进行分类。 <jupyter_code>from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences import numpy as np maxlen = 100 # We will cut reviews after 100 words(在 100 个单词后截断评论) training_samples = 200 # We will be training on 200 samples(在 200 个样本上训练) validation_samples = 10000 # We will be validating on 10000 samples(在 10 000 个样本上验证) max_words = 10000 # We will only consider the top 10,000 words in the dataset(只考虑数据集中前 10 000 个最常见的单词) tokenizer = Tokenizer(num_words=max_words) tokenizer.fit_on_texts(texts) sequences = tokenizer.texts_to_sequences(texts) word_index = tokenizer.word_index print('Found %s unique tokens.' % len(word_index)) data = pad_sequences(sequences, maxlen=maxlen) labels = np.asarray(labels) print('Shape of data tensor:', data.shape) print('Shape of label tensor:', labels.shape) # Split the data into a training set and a validation set # But first, shuffle the data, since we started from data # where sample are ordered (all negative first, then all positive). # 将数据划分为训练集和验证集,但首先要打乱数据,因为一开始数据中的样本是排好序的(所有负面评论都在前面, 然后是所有正面评论) indices = np.arange(data.shape[0]) np.random.shuffle(indices) data = data[indices] labels = labels[indices] x_train = data[:training_samples] y_train = labels[:training_samples] x_val = data[training_samples: training_samples + validation_samples] y_val = labels[training_samples: training_samples + validation_samples]<jupyter_output>Found 88582 unique tokens. Shape of data tensor: (25000, 100) Shape of label tensor: (25000,) <jupyter_text>### Download the GloVe word embeddings Head to `https://nlp.stanford.edu/projects/glove/` (where you can learn more about the GloVe algorithm), and download the pre-computed embeddings from 2014 English Wikipedia. It's a 822MB zip file named `glove.6B.zip`, containing 100-dimensional embedding vectors for 400,000 words (or non-word tokens). Un-zip it. ### 下载 GloVe 词嵌入 打开 https://nlp.stanford.edu/projects/glove ,下载 2014 年英文维基百科的预计算嵌入。这是 一个 822 MB 的压缩文件,文件名是 glove.6B.zip,里面包含 400 000 个单词(或非单词的标记) 的100 维嵌入向量。解压文件。 备注:我们下载了,文件放在 'data/glove.6B.100d.txt' ### Pre-process the embeddings Let's parse the un-zipped file (it's a `txt` file) to build an index mapping words (as strings) to their vector representation (as number vectors). ### 对嵌入进行预处理 我们对解压后的文件(一个 .txt 文件)进行解析,构建一个将单词(字符串)映射为其向量表示(数值向量)的索引。 <jupyter_code>glove_dir = 'data/' embeddings_index = {} f = open(os.path.join(glove_dir, 'glove.6B.100d.txt'),encoding='utf8') for line in f: values = line.split() word = values[0] coefs = np.asarray(values[1:], dtype='float32') embeddings_index[word] = coefs f.close() print('Found %s word vectors.' % len(embeddings_index))<jupyter_output>Found 400000 word vectors. <jupyter_text> Now let's build an embedding matrix that we will be able to load into an `Embedding` layer. It must be a matrix of shape `(max_words, embedding_dim)`, where each entry `i` contains the `embedding_dim`-dimensional vector for the word of index `i` in our reference word index (built during tokenization). Note that the index `0` is not supposed to stand for any word or token -- it's a placeholder. 接下来,需要构建一个可以加载到 Embedding 层中的嵌入矩阵。它必须是一个形状为(max_words, embedding_dim) 的矩阵,对于单词索引(在分词时构建)中索引为 i 的单词, 这个矩阵的元素 i 就是这个单词对应的 embedding_dim 维向量。注意,索引 0 不应该代表任何单词或标记,它只是一个占位符。 <jupyter_code>embedding_dim = 100 embedding_matrix = np.zeros((max_words, embedding_dim)) for word, i in word_index.items(): embedding_vector = embeddings_index.get(word) if i < max_words: if embedding_vector is not None: # Words not found in embedding index will be all-zeros. # 嵌入索引(embeddings_index) 中找不到的词,其嵌入向量全为 0 embedding_matrix[i] = embedding_vector<jupyter_output><empty_output><jupyter_text>### Define a model We will be using the same model architecture as before: ### 定义模型 我们将使用与前面相同的模型架构: <jupyter_code>from keras.models import Sequential from keras.layers import Embedding, Flatten, Dense model = Sequential() model.add(Embedding(max_words, embedding_dim, input_length=maxlen)) model.add(Flatten()) model.add(Dense(32, activation='relu')) model.add(Dense(1, activation='sigmoid')) model.summary()<jupyter_output>_________________________________________________________________ Layer (type) Output Shape Param # ================================================================= embedding_3 (Embedding) (None, 100, 100) 1000000 _________________________________________________________________ flatten_2 (Flatten) (None, 10000) 0 _________________________________________________________________ dense_2 (Dense) (None, 32) 320032 _________________________________________________________________ dense_3 (Dense) (None, 1) 33 ================================================================= Total params: 1,320,065 Trainable params: 1,320,065 Non-trainable params: 0 _________________________________________________________________ <jupyter_text>### Load the GloVe embeddings in the model The `Embedding` layer has a single weight matrix: a 2D float matrix where each entry `i` is the word vector meant to be associated with index `i`. Simple enough. Let's just load the GloVe matrix we prepared into our `Embedding` layer, the first layer in our model: ### 在模型中加载 GloVe 嵌入 Embedding 层只有一个权重矩阵,是一个二维的浮点数矩阵,其中每个元素 i 是与索引 i 相关联的词向量。够简单。将准备好的 GloVe 矩阵加载到 Embedding 层中,即模型的第一层。 <jupyter_code>model.layers[0].set_weights([embedding_matrix]) model.layers[0].trainable = False<jupyter_output><empty_output><jupyter_text> Additionally, we freeze the embedding layer (we set its `trainable` attribute to `False`), following the same rationale as what you are already familiar with in the context of pre-trained convnet features: when parts of a model are pre-trained (like our `Embedding` layer), and parts are randomly initialized (like our classifier), the pre-trained parts should not be updated during training to avoid forgetting what they already know. The large gradient update triggered by the randomly initialized layers would be very disruptive to the already learned features. 此外,需要冻结 Embedding 层(即将其 trainable 属性设为 False),其原理和预训练的卷积神经网络特征相同,你已经很熟悉了。如果一个模型的一部分是经过预训练的(如 Embedding 层),而另一部分是随机初始化的(如分类器),那么在训练期间不应该更新预训练的部分,以避免丢失它们所保存的信息。随机初始化的层会引起较大的梯度更新,会破坏已经学到的特征。### Train and evaluate Let's compile our model and train it: ### 训练模型与评估模型 编译并训练模型: <jupyter_code>model.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['acc']) history = model.fit(x_train, y_train, epochs=10, batch_size=32, validation_data=(x_val, y_val)) model.save_weights('pre_trained_glove_model.h5')<jupyter_output>Train on 200 samples, validate on 10000 samples Epoch 1/10 200/200 [==============================] - 1s 4ms/step - loss: 1.6337 - acc: 0.5250 - val_loss: 0.7130 - val_acc: 0.5100 Epoch 2/10 200/200 [==============================] - 1s 3ms/step - loss: 0.7565 - acc: 0.5800 - val_loss: 0.6910 - val_acc: 0.5418 Epoch 3/10 200/200 [==============================] - 1s 3ms/step - loss: 0.5956 - acc: 0.6950 - val_loss: 1.1205 - val_acc: 0.4936 Epoch 4/10 200/200 [==============================] - 1s 3ms/step - loss: 0.5335 - acc: 0.7350 - val_loss: 0.7134 - val_acc: 0.5362 Epoch 5/10 200/200 [==============================] - 1s 3ms/step - loss: 0.4713 - acc: 0.8100 - val_loss: 0.7177 - val_acc: 0.5589 Epoch 6/10 200/200 [==============================] - 0s 2ms/step - loss: 0.1448 - acc: 0.9800 - val_loss: 1.3373 - val_acc: 0.4952 Epoch 7/10 200/200 [==============================] - 1s 3ms/step - loss: 0.2545 - acc: 0.8800 - val_loss: 1.3110 - val_acc: 0.4960 Epoch 8/10 200/200 [========[...]<jupyter_text>Let's plot its performance over time: 绘制结果:<jupyter_code>import matplotlib.pyplot as plt acc = history.history['acc'] val_acc = history.history['val_acc'] loss = history.history['loss'] val_loss = history.history['val_loss'] epochs = range(1, len(acc) + 1) plt.plot(epochs, acc, 'bo', label='Training acc') plt.plot(epochs, val_acc, 'b', label='Validation acc') plt.title('Training and validation accuracy') plt.legend() plt.figure() plt.plot(epochs, loss, 'bo', label='Training loss') plt.plot(epochs, val_loss, 'b', label='Validation loss') plt.title('Training and validation loss') plt.legend() plt.show()<jupyter_output><empty_output><jupyter_text> The model quickly starts overfitting, unsurprisingly given the small number of training samples. Validation accuracy has high variance for the same reason, but seems to reach high 60s. Note that your mileage may vary: since we have so few training samples, performance is heavily dependent on which exact 200 samples we picked, and we picked them at random. If it worked really poorly for you, try picking a different random set of 200 samples, just for the sake of the exercise (in real life you don't get to pick your training data). We can also try to train the same model without loading the pre-trained word embeddings and without freezing the embedding layer. In that case, we would be learning a task-specific embedding of our input tokens, which is generally more powerful than pre-trained word embeddings when lots of data is available. However, in our case, we have only 200 training samples. Let's try it: 模型很快就开始过拟合,考虑到训练样本很少,这一点也不奇怪。出于同样的原因,验证精度的波动很大,但似乎达到了接近 60%。 注意,你的结果可能会有所不同。训练样本数太少,所以模型性能严重依赖于你选择的200个样本,而样本是随机选择的。如果你得到的结果很差,可以尝试重新选择 200 个不同的 随机样本,你可以将其作为练习(在现实生活中无法选择自己的训练数据)。 你也可以在不加载预训练词嵌入、也不冻结嵌入层的情况下训练相同的模型。在这种情况下, 你将会学到针对任务的输入标记的嵌入。如果有大量的可用数据,这种方法通常比预训练词嵌入更加强大,但本例只有 200 个训练样本。我们来试一下这种方法:。 <jupyter_code>from keras.models import Sequential from keras.layers import Embedding, Flatten, Dense model = Sequential() model.add(Embedding(max_words, embedding_dim, input_length=maxlen)) model.add(Flatten()) model.add(Dense(32, activation='relu')) model.add(Dense(1, activation='sigmoid')) model.summary() model.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['acc']) history = model.fit(x_train, y_train, epochs=10, batch_size=32, validation_data=(x_val, y_val)) acc = history.history['acc'] val_acc = history.history['val_acc'] loss = history.history['loss'] val_loss = history.history['val_loss'] epochs = range(1, len(acc) + 1) plt.plot(epochs, acc, 'bo', label='Training acc') plt.plot(epochs, val_acc, 'b', label='Validation acc') plt.title('Training and validation accuracy') plt.legend() plt.figure() plt.plot(epochs, loss, 'bo', label='Training loss') plt.plot(epochs, val_loss, 'b', label='Validation loss') plt.title('Training and validation loss') plt.legend() plt.show()<jupyter_output><empty_output><jupyter_text> Validation accuracy stalls in the low 50s. So in our case, pre-trained word embeddings does outperform jointly learned embeddings. If you increase the number of training samples, this will quickly stop being the case -- try it as an exercise. Finally, let's evaluate the model on the test data. First, we will need to tokenize the test data: 验证精度停留在 50% 多一点。因此,在本例中,预训练词嵌入的性能要优于与任务一起学习的嵌入。如果增加样本数量,情况将很快发生变化,你可以把它作为一个练习。 最后,我们在测试数据上评估模型。首先,你需要对测试数据进行分词。 <jupyter_code>test_dir = os.path.join(imdb_dir, 'test') labels = [] texts = [] for label_type in ['neg', 'pos']: dir_name = os.path.join(test_dir, label_type) for fname in sorted(os.listdir(dir_name)): if fname[-4:] == '.txt': f = open(os.path.join(dir_name, fname),encoding='utf8') texts.append(f.read()) f.close() if label_type == 'neg': labels.append(0) else: labels.append(1) sequences = tokenizer.texts_to_sequences(texts) x_test = pad_sequences(sequences, maxlen=maxlen) y_test = np.asarray(labels)<jupyter_output><empty_output><jupyter_text>And let's load and evaluate the first model: 接下来,加载并评估第一个模型。<jupyter_code>model.load_weights('pre_trained_glove_model.h5') model.evaluate(x_test, y_test)<jupyter_output>25000/25000 [==============================] - 1s 59us/step
permissive
/References/Python深度学习/code/6.1-using-word-embeddings.ipynb
Jianhan-Liu/NLP
14
<jupyter_start><jupyter_text><jupyter_code>import numpy as np import pandas as pd import matplotlib.pyplot as plt import seaborn as sns %matplotlib inline import os import zipfile !pip install -U -q PyDrive from pydrive.auth import GoogleAuth from pydrive.drive import GoogleDrive from google.colab import auth from oauth2client.client import GoogleCredentials import shutil ## GOOGLE DRIVE AUTHENTICATION auth.authenticate_user() gauth = GoogleAuth() gauth.credentials = GoogleCredentials.get_application_default() drive = GoogleDrive(gauth) ## SEACHING FOR THE ZIP FILE AND THEN UNZIPPING IT IN COLAB AND DELETING UN-NECESSARY FILES fid = drive.ListFile({'q':"title='train.csv'"}).GetList()[0]['id'] f = drive.CreateFile({'id': fid}) f.GetContentFile('train.csv') fid = drive.ListFile({'q':"title='test.csv'"}).GetList()[0]['id'] f = drive.CreateFile({'id': fid}) f.GetContentFile('test.csv') df = pd.read_csv('train.csv') df.head() test = pd.read_csv('test.csv') test.head() print(df.shape) df.describe() print(test.shape) test.describe() df.isna().sum().sort_values(ascending=False) test.isna().sum().sort_values(ascending=False)<jupyter_output><empty_output><jupyter_text>### Cleaning and analysing the datasetLet's remove few columns which does not look like to have importance and also few of the columns which have a lot of NaN's.<jupyter_code>df['belongs_to_collection'][0] df['homepage'][:5] ## tagline features seems important but there are too many NaN values(1/6th of dataset). df['tagline'][0] df['Keywords'][0] df['production_companies'][0] df['production_countries'][0] df['spoken_languages'][0] df['crew'][0] df['cast'][0] df['overview'][0] dropping_cols = ['belongs_to_collection','homepage','tagline'] df.drop(dropping_cols,axis=1,inplace=True) print(df.shape) df.head() dropping_cols = ['belongs_to_collection','homepage','tagline'] test.drop(dropping_cols,axis=1,inplace=True) print(test.shape) test.head() df.isna().sum().sort_values(ascending=False) test.isna().sum().sort_values(ascending=False) df.dropna(inplace=True) df.shape test.dropna(inplace=True) test.shape df.isna().sum().sort_values(ascending=False) test.isna().sum().sort_values(ascending=False)<jupyter_output><empty_output><jupyter_text>## Lets start to pre-process the dataset to use it for prediction<jupyter_code>sns.countplot(x='status',data=df) sns.boxplot(data=df,y='runtime') df.info() sns.jointplot(x='revenue',y='popularity',data=df) sns.jointplot(x='revenue',y='budget',data=df) sns.jointplot(x='revenue',y='runtime',data=df) plt.figure(figsize=(12,12)) sns.distplot(df['revenue'],rug=True) plt.figure(figsize=(13,20)) sns.countplot(y='original_language',data=df)<jupyter_output>/usr/local/lib/python3.6/dist-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use. stat_data = remove_na(group_data) <jupyter_text>Main language of movies are English.<jupyter_code>df['status'].value_counts() print(df['revenue'][df['status']=='Rumored']) ## applying log to revenue to normally distribute. df['logrevenue'] = np.log1p(df['revenue']) sns.distplot(df['logrevenue']) df['release_date'] = pd.to_datetime(df['release_date'], infer_datetime_format=True) df['release_day'] = df['release_date'].apply(lambda t: t.day) df['release_weekday'] = df['release_date'].apply(lambda t: t.weekday()) df['release_month'] = df['release_date'].apply(lambda t: t.month) # Year was being interpreted as future dates in some cases so I had to adjust some values df['release_year'] = df['release_date'].apply(lambda t: t.year if t.year < 2018 else t.year -100) df[['release_date','release_day','release_weekday','release_month','release_year']].head() test['release_date'] = pd.to_datetime(test['release_date'], infer_datetime_format=True) test['release_day'] = test['release_date'].apply(lambda t: t.day) test['release_weekday'] = test['release_date'].apply(lambda t: t.weekday()) test['release_month'] = test['release_date'].apply(lambda t: t.month) # Year was being interpreted as future dates in some cases so I had to adjust some values test['release_year'] = test['release_date'].apply(lambda t: t.year if t.year < 2018 else t.year -100) test[['release_date','release_day','release_weekday','release_month','release_year']].head() print(len(df[df['runtime']==0])) print(len(test[test['runtime']==0])) from collections import defaultdict def map_runtime(df): df['runtime'].fillna(0) run = df[(df['runtime'].notnull()) & (df['runtime'] != 0)] year_mean = run.groupby(['release_year'])['runtime'].agg('mean') d = dict(year_mean) for i in df[df['runtime'] == 0]: df['runtime'] = df.loc[:, 'release_year'].map(d) return df test = map_runtime(test) sns.countplot(x='release_month',data=df)<jupyter_output>/usr/local/lib/python3.6/dist-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use. stat_data = remove_na(group_data) <jupyter_text>Many movies come in the month of september and october<jupyter_code>plt.figure(figsize=(12,10)) sns.countplot(x='release_day',data=df)<jupyter_output>/usr/local/lib/python3.6/dist-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use. stat_data = remove_na(group_data) <jupyter_text>Most of the movies are either released on 1st of every month or 15th.<jupyter_code>plt.figure(figsize=(25,15)) sns.countplot(x='release_year',data=df) plt.xticks(rotation=90)<jupyter_output>/usr/local/lib/python3.6/dist-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use. stat_data = remove_na(group_data) <jupyter_text>Maximum number of movies released are in year 2013 to 2017<jupyter_code>sns.countplot(x='release_weekday',data=df)<jupyter_output>/usr/local/lib/python3.6/dist-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use. stat_data = remove_na(group_data) <jupyter_text>Most of the movies are released on Friday, Thrusday and Wednesday.<jupyter_code>df['genres'].describe() len(df[df['budget']==0]) import scipy.stats as stats X = df[df['budget'] != 0] for i in X.select_dtypes(include='number', exclude='datetime'): print(i, stats.pearsonr(X.budget, X[i])) # release_year and popularity correlate most strongly with budget def map_budget(df): d = defaultdict() #df['budget'] = df['budget'].fillna(0) X = df[df['budget'] != 0] year_mean = pd.Series(X.groupby(['release_year'])['budget'].agg('mean')) d = dict(year_mean) for i in df[df['budget'] == 0]: df['budget'] = df.loc[:, 'release_year'].map(d) # In a few cases, there are only 1 or 2 movies provided from a given year and are filled with Na values df.budget = df.sort_values(by='release_year').budget.fillna(method='ffill') return df df = map_budget(df) df.budget.describe() print(len(df[df['budget']==0])) len(test[test['budget']==0]) test = map_budget(test) len(test[test['budget']==0]) df.head() test.head() df['poster_path'].fillna(0, inplace=True) df.loc[df['poster_path'] != 0, 'poster_path'] = 1 test['poster_path'].fillna(0, inplace=True) test.loc[test['poster_path'] != 0, 'poster_path'] = 1 genres= df.genres.str.get_dummies(sep=',') df =pd.concat([df,genres],axis=1) df.drop(['genres'],axis=1,inplace=True) df.head() df.shape genres= test.genres.str.get_dummies(sep=',') test =pd.concat([test,genres],axis=1) test.drop(['genres'],axis=1,inplace=True) test.shape dropping_cols = ['id','imdb_id','original_title','overview','production_companies','status','title','Keywords','cast','crew','production_countries','release_date','spoken_languages'] df.drop(dropping_cols,axis=1,inplace=True) print(df.shape) df.head() dropping_cols = ['id','imdb_id','original_title','overview','production_companies','status','title','Keywords','cast','crew','production_countries','release_date','spoken_languages'] test.drop(dropping_cols,axis=1,inplace=True) print(test.shape) test.head() just_dummies = pd.get_dummies(df['original_language']) df = pd.concat([df, just_dummies], axis=1) df.drop(['original_language'], inplace=True, axis=1) print(df.shape) df.head() just_dummies = pd.get_dummies(test['original_language']) test = pd.concat([test, just_dummies], axis=1) test.drop(['original_language'], inplace=True, axis=1) print(test.shape) test.head() df.reset_index(inplace=True) test.reset_index(inplace=True) print(df.shape) df.head(5) print(test.shape) test.head(5)<jupyter_output>(3814, 121) <jupyter_text>## Building model for training and predictions <jupyter_code>from sklearn.model_selection import train_test_split from sklearn.metrics import classification_report X = df.drop(['revenue'],axis=1) y = df['revenue'] X_train,X_test, y_train, y_test = train_test_split(X,y,test_size=0.20,random_state=0) print("Shape of X_train is : ",X_train.shape[0]) print("Shape of y_train is : ",y_train.shape[0]) print("Shape of X_test is : ",X_test.shape[0]) print("Shape of y_test is : ",y_test.shape[0]) X_train.head() y_train.head() X_test.head() y_test.head() from sklearn.linear_model import LinearRegression from xgboost import XGBRegressor from lightgbm import LGBMRegressor lr_ = LinearRegression() lr_.fit(X_train,y_train) pred = lr_.predict(X_test) sns.distplot((y_test-pred),bins=50) plt.show() def rmsle(y_true, y_pred): return 'rmsle', np.sqrt(np.mean(np.power(np.log1p(y_pred) - np.log1p(y_true), 2))), False from sklearn import metrics print('MAE:', metrics.mean_absolute_error(y_test, pred)) print('MSE:', metrics.mean_squared_error(y_test, pred)) print('RMSE:', np.sqrt(metrics.mean_squared_error(y_test, pred))) print('RMSLE:', rmsle(y_test, pred)) lr = LGBMRegressor(boosting_type='dart',num_leaves=20,max_depth=-1,min_data_in_leaf=20, learning_rate=0.2,n_estimators=500,subsample_for_bin=200000, class_weight=None,min_split_gain=0.0,min_child_weight=0.001,subsample=0.1,subsample_freq=0,colsample_bytree=0.75,reg_alpha=0.0,reg_lambda=0.0, random_state=101,n_jobs=-1) lr.fit(X_train, y_train,eval_set=[(X_test, y_test)],eval_metric=rmsle,verbose=False) pred = lr.predict(X_test, num_iteration=lr.best_iteration_) print('MAE:', metrics.mean_absolute_error(y_test, pred)) print('MSE:', metrics.mean_squared_error(y_test, pred)) print('RMSE:', np.sqrt(metrics.mean_squared_error(y_test, pred))) print('RMSLE:', rmsle(y_test, pred)) fig, ax = plt.subplots() ax.scatter(y_test, pred) ax.plot([y_test.min(), y_test.max()], [y_test.min(), y_test.max()], lw=1) ax.set_xlabel('Measured') ax.set_ylabel('Predicted') plt.show() lr.best_score_ # features = list(X.columns) importance = lr.feature_importances_ # feature_indexes_by_importance = importance.argsort() # for index in feature_indexes_by_importance: # print('{}-{:.2f}%'.format(features[index], (importance[index] *100.0))) importance lr.n_estimators<jupyter_output><empty_output>
no_license
/TMDB_Box_Office_Revenue_Prediction.ipynb
inboxpraveen/TMDB-Box-office-revenue
9
<jupyter_start><jupyter_text># Data Scaling ## StandardScaler (Standardization) y = (x – mean) / standard_deviation Where the mean is calculated as: mean = sum(x) / count(x) And the standard_deviation is calculated as: standard_deviation = sqrt( sum( (x – mean)^2 ) / count(x)) <jupyter_code>from sklearn.preprocessing import StandardScaler data = [[0, 0], [0, 0], [1, 1], [1, 1]] scaler = StandardScaler() scaler.fit(data) print(scaler.mean_) newdata = scaler.transform(data) print(newdata) # or newdata = scaler.fit_transform(data) print(newdata)<jupyter_output>[[-1. -1.] [-1. -1.] [ 1. 1.] [ 1. 1.]] <jupyter_text>using in train and test data just example without data sc_X = StandardScaler() X_train = sc_X.fit_transform(X_train) X_test = sc_X.transform(X_test) sc_y = StandardScaler() y_train = sc_y.fit_transform(y_train) y_test = sc_y.fit_transform(y_test) ## MinMaxScaler (Normalization) y = (x – min) / (max – min) Where the minimum and maximum values pertain to the value x being normalized <jupyter_code>from sklearn.preprocessing import MinMaxScaler data = [[-1, 2], [-0.5, 6], [0, 10], [1, 18]] scaler = MinMaxScaler() scaler.fit(data) print(scaler.data_range_) print(scaler.data_min_) print(scaler.data_max_) newdata = scaler.transform(data) print(newdata)<jupyter_output>[ 2. 16.] [-1. 2.] [ 1. 18.] [[0. 0. ] [0.25 0.25] [0.5 0.5 ] [1. 1. ]] <jupyter_text>## Normalizer و هي مخصصة لتناول كل صف علي حدة في المصفوفات ثنائية الأبعاد<jupyter_code>from sklearn.preprocessing import Normalizer X = [[4, 1, 2, 2], [1, 3, 9, 3], [5, 7, 5, 1]] # تستخدم l1 لجعل مجموع كل صف هو القيمة العظمي #transformer = Normalizer(norm='l1' ) # تستخدم l2 لجعل جذر مجموع مربعات كل صف هو القيمة العظمي #transformer = Normalizer(norm='l2' ) # تستخدم max لجعل القيمة العظمي في كل صف هي القيمة العظمي transformer = Normalizer(norm='max' ) transformer.fit(X) transformer.transform(X)<jupyter_output><empty_output><jupyter_text>## MaxAbsScaler مشابهة لـ normalizer لكن بالنسبة للعمود و ليس الصف حيث تجعل أكبر قيمة في كل عمود هي القيمة العظمي و تغير الباقيين علي اساسه<jupyter_code>from sklearn.preprocessing import MaxAbsScaler X = [[ 1., 10., 2.], [ 2., 0., 0.], [ 5., 1., -1.]] transformer = MaxAbsScaler().fit(X) transformer transformer.transform(X)<jupyter_output><empty_output><jupyter_text>## FunctionTransformer وهي تستخدم لعمل سكيل بدالة اقوم كتابتها بنفسي FunctionTransformer(func=None, inverse_func=None, validate= None, accept_sparse=False,pass_y='deprecated', check_inverse=True, kw_args=None,inv_kw_args=None)<jupyter_code>import numpy as np from sklearn.preprocessing import FunctionTransformer X = [[4, 1, 2, 2], [1, 3, 9, 3], [5, 7, 5, 1]] def function1(z): return np.sqrt(z) FT = FunctionTransformer(func = function1) FT.fit(X) newdata = FT.transform(X) newdata<jupyter_output><empty_output><jupyter_text>## Binarizer و هي تقوم بتحويل كل الأرقام إلي صفر او واحد , حسب قيمة العتبة threshold<jupyter_code>from sklearn.preprocessing import Binarizer X = [[ 1., -1., -2.],[ 2., 0., -1.], [ 0., 1., -1.]] transformer = Binarizer(threshold=1.5 ) transformer.fit(X) transformer transformer.transform(X)<jupyter_output><empty_output><jupyter_text>## PolynomialFeatures وهي خاصة بعمل فيتشرز جديدة , هي عبارة عن حاصل ضرب الفيتشرز الحالية بطريقة البولونوميال , فلو كانت الدرجة 2 مثلا , و كان لدينا اصلا عمودين فقط اي 2 فيتشرز , فسيعطي لنا علي الترتيب<jupyter_code>import numpy as np from sklearn.preprocessing import PolynomialFeatures X = np.arange(6).reshape(3, 2) print(X) # يتم كتابة الدرجة , وهل تحتوي علي قيمة بياس (رقم 1) ام لا poly = PolynomialFeatures(degree=2 , include_bias = True) poly.fit_transform(X) # ولو تم اختيار interaction_only كقيمة True سيعرض فقط قيم a مضروبة في b و يحذف الاسس للقيم الوحيدة poly = PolynomialFeatures(interaction_only=True) poly.fit_transform(X) <jupyter_output><empty_output>
no_license
/Data_Scaling.ipynb
abohashem95/SKlearn-codes
7
<jupyter_start><jupyter_text># Lab 4: Functions and VisualizationsWelcome to Lab 4! This week, we'll learn about functions, table methods such as `apply`, and how to generate visualizations! Recommended Reading: * [Applying a Function to a Column](https://www.inferentialthinking.com/chapters/08/1/applying-a-function-to-a-column.html) * [Visualizations](https://www.inferentialthinking.com/chapters/07/visualization.html) First, set up the notebook by running the cell below.<jupyter_code>import numpy as np from datascience import * # These lines set up graphing capabilities. import matplotlib %matplotlib inline import matplotlib.pyplot as plt plt.style.use('fivethirtyeight') import warnings warnings.simplefilter('ignore', FutureWarning) from ipywidgets import interact, interactive, fixed, interact_manual import ipywidgets as widgets # When you log-in please hit return (not shift + return) after typing in your email from client.api.notebook import Notebook ok = Notebook('lab04.ok') _ = ok.submit()<jupyter_output>===================================================================== Assignment: Functions and Visualizations OK, version v1.18.1 ===================================================================== <jupyter_text>## 1. Defining functions Let's write a very simple function that converts a proportion to a percentage by multiplying it by 100. For example, the value of `to_percentage(.5)` should be the number 50 (no percent sign). A function definition has a few parts. ##### `def` It always starts with `def` (short for **def**ine): def ##### Name Next comes the name of the function. Like other names we've defined, it can't start with a number or contain spaces. Let's call our function `to_percentage`: def to_percentage ##### Signature Next comes something called the *signature* of the function. This tells Python how many arguments your function should have, and what names you'll use to refer to those arguments in the function's code. A function can have any number of arguments (including 0!). `to_percentage` should take one argument, and we'll call that argument `proportion` since it should be a proportion. def to_percentage(proportion) If we want our function to take more than one argument, we add a comma between each argument name. Note that if we had zero arguments, we'd still place the parentheses () after that name. We put a **colon** after the signature to tell Python that the next indented lines are the body of the function. If you're getting a syntax error after defining a function, check to make sure you remembered the colon! def to_percentage(proportion): ##### Documentation Functions can do complicated things, so you should write an explanation of what your function does. For small functions, this is less important, but it's a good habit to learn from the start. Conventionally, Python functions are documented by writing an **indented** triple-quoted string: def to_percentage(proportion): """Converts a proportion to a percentage.""" ##### Body Now we start writing code that runs when the function is called. This is called the *body* of the function and every line **must be indented with a tab**. Any lines that are *not* indented and left-aligned with the def statement is considered outside the function. Some notes about the body of the function: - We can write code that we would write anywhere else. - We use the arguments defined in the function signature. We can do this because we assume that when we call the function, values are already assigned to those arguments. - We generally avoid referencing variables defined *outside* the function. If you would like to reference variables outside of the function, pass them through as arguments! Now, let's give a name to the number we multiply a proportion by to get a percentage: def to_percentage(proportion): """Converts a proportion to a percentage.""" factor = 100 ##### `return` The special instruction `return` is part of the function's body and tells Python to make the value of the function call equal to whatever comes right after `return`. We want the value of `to_percentage(.5)` to be the proportion .5 times the factor 100, so we write: def to_percentage(proportion): """Converts a proportion to a percentage.""" factor = 100 return proportion * factor `return` only makes sense in the context of a function, and **can never be used outside of a function**. `return` is always the last line of the function because Python stops executing the body of a function once it hits a `return` statement. If a function does not have a return statement, it will not return anything; if you expect a value back from the function, make sure to include a return statement. *Note:* `return` inside a function tells Python what value the function evaluates to. However, there are other functions, like `print`, that have no `return` value. For example, `print` simply prints a certain value out to the console. `return` and `print` are **very** different. **Question 1.1.** Define `to_percentage` in the cell below. Call your function to convert the proportion .2 to a percentage. Name that percentage `twenty_percent`. <!-- BEGIN QUESTION name: q11 --><jupyter_code>def ... '''...''' ... = ... return ... twenty_percent = ... twenty_percent ok.grade("q11");<jupyter_output><empty_output><jupyter_text>Like you’ve done with built-in functions in previous labs (max, abs, etc.), you can pass in named values as arguments to your function. **Question 1.2.** Use `to_percentage` again to convert the proportion named `a_proportion` (defined below) to a percentage called `a_percentage`. *Note:* You don't need to define `to_percentage` again! Like other named values, functions stick around after you define them. <!-- BEGIN QUESTION name: q12 --><jupyter_code>a_proportion = 2**(.5) / 2 a_percentage = ... a_percentage ok.grade("q12");<jupyter_output><empty_output><jupyter_text>Here's something important about functions: the names assigned *within* a function body are only accessible within the function body. Once the function has returned, those names are gone. So even if you created a variable called `factor` and defined `factor = 100` inside of the body of the `to_percentage` function and then called `to_percentage`, `factor` would not have a value assigned to it outside of the body of `to_percentage`:<jupyter_code># You should see an error when you run this. (If you don't, you might # have defined factor somewhere above.) factor<jupyter_output><empty_output><jupyter_text>As we've seen with built-in functions, functions can also take strings (or arrays, or tables) as arguments, and they can return those things, too. **Question 1.3.** Define a function called `disemvowel`. It should take a single string as its argument. (You can call that argument whatever you want.) It should return a copy of that string, but with all the characters that are vowels removed. (In English, the vowels are the characters "a", "e", "i", "o", and "u".) You can use as many lines inside of the function to do this as you’d like. *Hint:* To remove all the "a"s from a string, you can use `a_string.replace("a", "")`. The `.replace` method for strings returns a new string, so you can call `replace` multiple times, one after the other. <!-- BEGIN QUESTION name: q13 --><jupyter_code>def disemvowel(a_string): ... ... # An example call to your function. (It's often helpful to run # an example call from time to time while you're writing a function, # to see how it currently works.) disemvowel("Can you read this without vowels?") ok.grade("q13");<jupyter_output><empty_output><jupyter_text>##### Calls on calls on calls Just as you write a series of lines to build up a complex computation, it's useful to define a series of small functions that build on each other. Since you can write any code inside a function's body, you can call other functions you've written. If a function is a like a recipe, defining a function in terms of other functions is like having a recipe for cake telling you to follow another recipe to make the frosting, and another to make the jam filling. This makes the cake recipe shorter and clearer, and it avoids having a bunch of duplicated frosting recipes. It's a foundation of productive programming. For example, suppose you want to count the number of characters *that aren't vowels* in a piece of text. One way to do that is this to remove all the vowels and count the size of the remaining string. **Question 1.4.** Write a function called `num_non_vowels`. It should take a string as its argument and return a number. That number should be the number of characters in the argument string that aren't vowels. You should use the `disemvowel` function you wrote above inside of the `num_non_vowels` function. *Hint:* The function `len` takes a string as its argument and returns the number of characters in it. <!-- BEGIN QUESTION name: q14 --><jupyter_code>def num_non_vowels(a_string): """The number of characters in a string, minus the vowels.""" ... # Try calling your function yourself to make sure the output is what # you expect. ok.grade("q14");<jupyter_output><empty_output><jupyter_text>Functions can also encapsulate code that *displays output* instead of computing a value. For example, if you call `print` inside a function, and then call that function, something will get printed. The `movies_by_year` dataset in the textbook has information about movie sales from 1980 to 2015. Suppose you'd like to display the year with the 5th-highest total gross movie sales, printed in a human-readable way. You might do this:<jupyter_code>movies_by_year = Table.read_table("movies_by_year.csv") rank = 5 fifth_from_top_movie_year = movies_by_year.sort("Total Gross", descending=True).column("Year").item(rank-1) print("Year number", rank, "for total gross movie sales was:", fifth_from_top_movie_year)<jupyter_output><empty_output><jupyter_text>After writing this, you realize you also wanted to print out the 2nd and 3rd-highest years. Instead of copying your code, you decide to put it in a function. Since the rank varies, you make that an argument to your function. **Question 1.5.** Write a function called `print_kth_top_movie_year`. It should take a single argument, the rank of the year (like 2, 3, or 5 in the above examples). It should print out a message like the one above. *Note:* Your function shouldn't have a `return` statement. <!-- BEGIN QUESTION name: q15 --><jupyter_code>def print_kth_top_movie_year(k): ... ... print(...) # Example calls to your function: print_kth_top_movie_year(2) print_kth_top_movie_year(3) ok.grade("q15"); # interact also allows you to pass in an array for a function argument. It will # then present a dropdown menu of options. _ = interact(print_kth_top_movie_year, k=np.arange(1, 10))<jupyter_output><empty_output><jupyter_text>### `print` is not the same as `return` The `print_kth_top_movie_year(k)` function prints the total gross movie sales for the year that was provided! However, since we did not return any value in this function, we can not use it after we call it. Let's look at an example of another function that prints a value but does not return it.<jupyter_code>def print_number_five(): print(5) print_number_five()<jupyter_output><empty_output><jupyter_text>However, if we try to use the output of `print_number_five()`, we see that the value `5` is printed but we get a TypeError when we try to add the number 2 to it!<jupyter_code>print_number_five_output = print_number_five() print_number_five_output + 2<jupyter_output><empty_output><jupyter_text>It may seem that `print_number_five()` is returning a value, 5. In reality, it just displays the number 5 to you without giving you the actual value! If your function prints out a value without returning it and you try to use that value, you will run into errors, so be careful! Explain to your neighbor how you might add a line of code to the `print_number_five` function (after `print(5)`) so that the code `print_number_five_output + 5` would result in the value `10`, rather than an error.## 2. Functions and CEO Incomes In this question, we'll look at the 2015 compensation of CEOs at the 100 largest companies in California. The data was compiled from a [Los Angeles Times analysis](http://spreadsheets.latimes.com/california-ceo-compensation/), and ultimately came from [filings](https://www.sec.gov/answers/proxyhtf.htm) mandated by the SEC from all publicly-traded companies. Two companies have two CEOs, so there are 102 CEOs in the dataset. We've copied the raw data from the LA Times page into a file called `raw_compensation.csv`. (The page notes that all dollar amounts are in **millions of dollars**.)<jupyter_code>raw_compensation = Table.read_table('raw_compensation.csv') raw_compensation<jupyter_output><empty_output><jupyter_text>We want to compute the average of the CEOs' pay. Try running the cell below.<jupyter_code>np.average(raw_compensation.column("Total Pay"))<jupyter_output><empty_output><jupyter_text>You should see a TypeError. Let's examine why this error occurred by looking at the values in the `Total Pay` column. **Question 2.1.** Use the `type` function and set `total_pay_type` to the type of the first value in the "Total Pay" column. <!-- BEGIN QUESTION name: q21 --><jupyter_code>total_pay_type = ... total_pay_type ok.grade("q21");<jupyter_output><empty_output><jupyter_text>**Question 2.2.** You should have found that the values in the `Total Pay` column are strings. It doesn't make sense to take the average of string values, so we need to convert them to numbers if we want to do this. Extract the first value in `Total Pay`. It's Mark Hurd's pay in 2015, in *millions* of dollars. Call it `mark_hurd_pay_string`. <!-- BEGIN QUESTION name: q22 --><jupyter_code>mark_hurd_pay_string = ... mark_hurd_pay_string ok.grade("q22");<jupyter_output><empty_output><jupyter_text>**Question 2.3.** Convert `mark_hurd_pay_string` to a number of *dollars*. Some hints, as this question requires multiple steps: - The string method `strip` will be useful for removing the dollar sign; it removes a specified character from the start or end of a string. For example, the value of `"100%".strip("%")` is the string `"100"`. - You'll also need the function `float`, which converts a string that looks like a number to an actual number. - Finally, remember that the answer should be in dollars, not millions of dollars. <!-- BEGIN QUESTION name: q23 --><jupyter_code>mark_hurd_pay = ... mark_hurd_pay ok.grade("q23");<jupyter_output><empty_output><jupyter_text>To compute the average pay, we need to do this for every CEO. But that looks like it would involve copying this code 102 times. This is where functions come in. First, we'll define a new function, giving a name to the expression that converts "total pay" strings to numeric values. Later in this lab, we'll see the payoff: we can call that function on every pay string in the dataset at once. The next section of this lab explains how to define a function For now, just fill in the ellipses in the cell below. **Question 2.4.** Copy the expression you used to compute `mark_hurd_pay`, and use it as the return expression of the function below. But make sure you replace the specific `mark_hurd_pay_string` with the generic `pay_string` name specified in the first line in the `def` statement. *Hint*: When dealing with functions, you should generally not be referencing any variable outside of the function. Usually, you want to be working with the arguments that are passed into it, such as `pay_string` for this function. If you're using `mark_hurd_pay_string` within your function, you're referencing an outside variable! <!-- BEGIN QUESTION name: q24 --><jupyter_code>def convert_pay_string_to_number(pay_string): """Converts a pay string like '$100' (in millions) to a number of dollars.""" ... ok.grade("q24");<jupyter_output><empty_output><jupyter_text>Running that cell doesn't convert any particular pay string. Instead, it creates a function called `convert_pay_string_to_number` that can convert *any* string with the right format to a number representing millions of dollars. We can call our function just like we call the built-in functions we've seen. It takes one argument -- a string -- and it returns a float.<jupyter_code>convert_pay_string_to_number('$42') convert_pay_string_to_number(mark_hurd_pay_string) # We can also compute Safra Catz's pay in the same way: convert_pay_string_to_number(raw_compensation.where("Name", are.containing("Safra")).column("Total Pay").item(0))<jupyter_output><empty_output><jupyter_text>So, what have we gained by defining the `convert_pay_string_to_number` function? Well, without it, we'd have to copy the code `10**6 * float(some_pay_string.strip("$"))` each time we wanted to convert a pay string. Now we just call a function whose name says exactly what it's doing.## 3. `apply`ing functions Defining a function is a lot like giving a name to a value with `=`. In fact, a function is a value just like the number 1 or the text "data"! For example, we can make a new name for the built-in function `max` if we want:<jupyter_code>our_name_for_max = max our_name_for_max(2, 6)<jupyter_output><empty_output><jupyter_text>The old name for `max` is still around:<jupyter_code>max(2, 6)<jupyter_output><empty_output><jupyter_text>Try just writing `max` or `our_name_for_max` (or the name of any other function) in a cell, and run that cell. Python will print out a (very brief) description of the function.<jupyter_code>max<jupyter_output><empty_output><jupyter_text>Now try writing `?max` or `?our_name_for_max` (or the name of any other function) in a cell, and run that cell. A information box should show up at the bottom of your screen a longer description of the function *Note: You can also press Shift+Tab after clicking on a name to see similar information!*<jupyter_code>?our_name_for_max<jupyter_output><empty_output><jupyter_text>Let's look at what happens when we set `max`to a non-function value. You'll notice that a TypeError will occur when you try calling `max`. Things like integers and strings are not callable. Look out for any functions that might have been renamed when you encounter this type of error<jupyter_code>max = 6 max(2, 6) # This cell resets max to the built-in function. Just run this cell, don't change its contents import builtins max = builtins.max<jupyter_output><empty_output><jupyter_text>Why is this useful? Since functions are just values, it's possible to pass them as arguments to other functions. Here's a simple but not-so-practical example: we can make an array of functions.<jupyter_code>make_array(max, np.average, are.equal_to)<jupyter_output><empty_output><jupyter_text>**Question 3.1.** Make an array containing any 3 other functions you've seen. Call it `some_functions`. <!-- BEGIN QUESTION name: q31 --><jupyter_code>some_functions = ... some_functions ok.grade("q31");<jupyter_output><empty_output><jupyter_text>Working with functions as values can lead to some funny-looking code. For example, see if you can figure out why the following code works. Check your explanation with a neighbor.<jupyter_code>make_array(max, np.average, are.equal_to).item(0)(4, -2, 7)<jupyter_output><empty_output><jupyter_text>A more useful example of passing functions to other functions as arguments is the table method `apply`. `apply` calls a function many times, once on *each* element in a column of a table. It produces an *array* of the results. Here we use `apply` to convert every CEO's pay to a number, using the function you defined:<jupyter_code>raw_compensation.apply(convert_pay_string_to_number, "Total Pay")<jupyter_output><empty_output><jupyter_text>Here's an illustration of what that did: Note that we didn’t write `raw_compensation.apply(convert_pay_string_to_number(), “Total Pay”)` or `raw_compensation.apply(convert_pay_string_to_number(“Total Pay”))`. We just passed the name of the function, with no parentheses, to `apply`, because all we want to do is let `apply` know the name of the function we’d like to use and the name of the column we’d like to use it on. `apply` will then call the function `convert_pay_string_to_number` on each value in the column for us! **Question 3.2.** Using `apply`, make a table that's a copy of `raw_compensation` with one additional column called `Total Pay ($)`. That column should contain the result of applying `convert_pay_string_to_number` to the `Total Pay` column (as we did above). Call the new table `compensation`. <!-- BEGIN QUESTION name: q32 --><jupyter_code>compensation = raw_compensation.with_column( "Total Pay ($)", ... ) compensation ok.grade("q32");<jupyter_output><empty_output><jupyter_text>Now that we have all the pays as numbers, we can learn more about them through computation. **Question 3.3.** Compute the average total pay of the CEOs in the dataset. <!-- BEGIN QUESTION name: q33 --><jupyter_code>average_total_pay = ... average_total_pay ok.grade("q33");<jupyter_output><empty_output><jupyter_text>**Question 3.4.** Companies pay executives in a variety of ways: in cash, by granting stock or other equity in the company, or with ancillary benefits (like private jets). Compute the proportion of each CEO's pay that was cash. (Your answer should be an array of numbers, one for each CEO in the dataset.) *Note:* When you answer this question, you'll encounter a red box appearing below your code cell that says something like `RuntimeWarning: invalid value encountered in true_divide`. Don't worry too much about the message. Warnings are raised by Python when it encounters an unusual condition in your code, but the condition is not severe enough to warrant throwing an error. The warning below is Python's cryptic way of telling you that you're dividing a number by zero. If you extract the values in `Total Pay ($)` as an array, you'll see that the last element is 0. <!-- BEGIN QUESTION name: q34 --><jupyter_code>cash_proportion = ... cash_proportion ok.grade("q34");<jupyter_output><empty_output><jupyter_text>Check out the `% Change` column in `compensation`. It shows the percentage increase in the CEO's pay from the previous year. For CEOs with no previous year on record, it instead says "(No previous year)". The values in this column are *strings*, not numbers, so like the `Total Pay` column, it's not usable without a bit of extra work. Given your current pay and the percentage increase from the previous year, you can compute your previous year's pay. For example, if your pay is $\$120$ this year, and that's an increase of 50% from the previous year, then your previous year's pay was $\frac{\$120}{1 + \frac{50}{100}}$, or \$80. **Question 3.5.** Create a new table called `with_previous_compensation`. It should be a copy of `compensation`, but with the "(No previous year)" CEOs filtered out, and with an extra column called `2014 Total Pay ($)`. That column should have each CEO's pay in 2014. *Hint 1:* You can print out your results after each step to make sure you're on the right track. *Hint 2:* We've provided a structure that you can use to get to the answer. However, if it's confusing, feel free to delete the current structure and approach the problem your own way! <!-- BEGIN QUESTION name: q35 --><jupyter_code># Definition to turn percent to number def percent_string_to_num(percent_string): """Converts a percentage string to a number.""" return ... # Compensation table where there is a previous year having_previous_year = ... # Get the percent changes as numbers instead of strings # We're still working off the table having_previous_year percent_changes = ... # Calculate the previous year's pay # We're still working off the table having_previous_year previous_pay = ... # Put the previous pay column into the having_previous_year table with_previous_compensation = ... with_previous_compensation ok.grade("q35");<jupyter_output><empty_output><jupyter_text>**Question 3.6.** What was the average pay of these CEOs in 2014? <!-- BEGIN QUESTION name: q36 --><jupyter_code>average_pay_2014 = ... average_pay_2014 ok.grade("q36");<jupyter_output><empty_output><jupyter_text>**Why is `apply` useful?** For operations like arithmetic, or the functions in the NumPy library, you don't need to use `apply`, because they automatically work on each element of an array. But there are many things that don't. The string manipulation we did in today's lab is one example. Since you can write any code you want in a function, `apply` gives you total control over how you operate on data.## 4. Histograms Earlier, we computed the average pay among the CEOs in our 102-CEO dataset. The average doesn't tell us everything about the amounts CEOs are paid, though. Maybe just a few CEOs make the bulk of the money, even among these 102. We can use a *histogram* method to display the *distribution* of a set of numbers. The table method `hist` takes a single argument, the name of a column of numbers. It produces a histogram of the numbers in that column. **Question 4.1.** Make a histogram of the total pay of the CEOs in `compensation`. Check with your neighbor or professor to make sure you have the right plot.<jupyter_code>...<jupyter_output><empty_output><jupyter_text>**Question 4.2.** How many CEOs made more than $30 million in total pay? Find the value using code, then check that the value you found is consistent with what you see in the histogram. *Hint:* Use the table method `where` and the property `num_rows`. <!-- BEGIN QUESTION name: q42 --><jupyter_code>num_ceos_more_than_30_million_2 = ... num_ceos_more_than_30_million_2 ok.grade("q42");<jupyter_output><empty_output><jupyter_text>Great job! You're finished with lab 4! Be sure to... * **run all the tests** (the next cell has a shortcut for that), * **Save and Checkpoint** from the File menu, * **run the last cell to submit your work**,<jupyter_code># For your convenience, you can run this cell to run all the tests at once! import os _ = [ok.grade(q[:-3]) for q in os.listdir("tests") if q.startswith('q')] _ = ok.submit()<jupyter_output><empty_output>
no_license
/lab04/lab04.ipynb
Peter-Jantsch/m121-sp21-lab
34
<jupyter_start><jupyter_text># List<jupyter_code># make a list with integer, float, string types and store it to var var = [10 ,"Apple" ,"a+b", 5.5, "Ball" ] # access individual items of list using square brackets [] in variable name print(var[0]) print(var[1]) print(var[2]) # use negative indexing print(var[-1]) print(var[-2]) # use len() to find length brands = ["Coke", "Apple", "Google", "Microsoft", "Toyota"] num_brands = len(brands) print(num_brands) # Slice a list in python using positive and negative indexing fruits = ["Apple", "Banana", "Mango", "Grapes", "Orange"] print(fruits[1:4]) print(fruits[ : 3]) print(fruits[-4:]) print(fruits[-3:-1]) # Modify elements of a list in python using index of the elements fruits = ["Apple", "Banana", "Mango", "Grapes", "Orange"] fruits[1] = "Pineapple" fruits[-1] = "Guava" print(fruits) # Combine two arrays using + operator concat = [1, 2, 3] concat + [4,5,6] print(concat) # Repeat elements in a list using * operator repeat = ["a"] repeat = repeat * 5 print(repeat) # Add an element in a list using append() add = ['a', 'b', 'c'] add.append('d') print(add) # Remove elements of an array using del, remove() and pop() colors = ["violet", "indigo", "blue", "green", "yellow", "orange", "red" ] del color[4] colors.remove("blue") colors.pop(3) print(color) # Declare a two-dimensional list in python using lists and accessing it using list index multd = [[1,2], [3,4], [5,6], [7,8]] print(multd[0]) print(multd[3]) print(multd[2][1]) print(multd[3][0])<jupyter_output><empty_output>
no_license
/python_list.ipynb
zhubaiyuan/exercise-book
1
<jupyter_start><jupyter_text># Implement a Queue - Using Two Stacks Given the Stack class below, implement a Queue class using **two** stacks! Note, this is a "classic" interview problem. Use a Python list data structure as your Stack.<jupyter_code># Uses lists instead of your own Stack class. stack1 = [] stack2 = []<jupyter_output><empty_output><jupyter_text>## Solution Fill out your solution below:<jupyter_code>class Queue2Stacks(object): def __init__(self): # Two Stacks self.instack = [] self.outstack = [] def enqueue(self,element): self.instack.append(element) def dequeue(self): if not self.outstack: while self.instack: self.outstack.append(self.instack.pop()) return self.outstack.pop() class MyQueue: def __init__(self): """ Initialize your data structure here. """ self.instack = [] self.outstack = [] def __repr__(self): return('instack {0}, outstack {1}'.format(str(self.instack), \ str(self.outstack))) def push(self, x): """ Push element x to the back of queue. :type x: int :rtype: void """ self.instack.append(x) def pop(self): """ Removes the element from in front of queue and returns that element. :rtype: int """ self.peek() return self.outstack.pop() def peek(self): """ Get the front element. :rtype: int """ if self.outstack == []: while self.instack != []: self.outstack.append(self.instack.pop()) last_peek = self.outstack[-1] return last_peek def empty(self): """ Returns whether the queue is empty. :rtype: bool """ return not self.instack and not self.outstack Testqueue = MyQueue() Testqueue.push(1) print(Testqueue) Testqueue.push(2) print(Testqueue) Testqueue.peek() print(Testqueue) Testqueue.pop() print(Testqueue) Testqueue.peek() print(Testqueue) Testqueue.pop() print(Testqueue) Testqueue.empty()<jupyter_output>instack [1], outstack [] instack [1, 2], outstack [] instack [], outstack [2, 1] instack [], outstack [2] instack [], outstack [2] instack [], outstack [] <jupyter_text># Test Your Solution You should be able to tell with your current knowledge of Stacks and Queues if this is working as it should. For example, the following should print as such:<jupyter_code>""" RUN THIS CELL TO CHECK THAT YOUR SOLUTION OUTPUT MAKES SENSE AND BEHAVES AS A QUEUE """ q = Queue2Stacks() for i in range(5): q.enqueue(i) for i in range(5): print(q.dequeue())<jupyter_output>0 1 2 3 4
no_license
/Python_algo/Chapter13_QueuesDeques/Stacks, Queues, and Deques Interview Problems/Stacks, Queues, Deques Interview Questions/.ipynb_checkpoints/Implement a Queue -Using Two Stacks -checkpoint.ipynb
qy2205/MyUdemy
3
<jupyter_start><jupyter_text>#### for improving model performance we need to split all wines types in three category good,fine and bad<jupyter_code>reviews = [] for i in data['quality']: if i >= 1 and i <= 3: reviews.append('1') elif i >= 4 and i <= 7: reviews.append('2') elif i >= 8 and i <= 10: reviews.append('3') data['Reviews'] = reviews data.head()<jupyter_output><empty_output><jupyter_text>### Splitting Data x = independent values y = depended values review is the only one depended yea it's review **notice: you can select last index into coulmn index by : -1**<jupyter_code>x = data.iloc[:, 0:-2].values y = data.iloc[:, -1].values<jupyter_output><empty_output><jupyter_text>### Creatting the Model here we create pip line with Standard scaler step(1) and KernelPCA step(2) and logistic regression model step(3) we will change it later any way ### A-Dimensionality reduction you can read more about it here [Principal component](https://en.wikipedia.org/wiki/Principal_component_analysis) , but simply you can reduce your Dimensions easly , the dimensions is your columns count and this process try to reduce it with out effect with your data. PCA , PCA KERNEL and plotting explained_variance to see the best components number for setting !<jupyter_code>sc_x=StandardScaler() x = sc_x.fit_transform(x) pca=PCA() x_pca = pca.fit_transform(x) plt.figure(figsize=(10,10)) plt.plot(np.cumsum(pca.explained_variance_ratio_), 'ro-') plt.grid() pca_new = PCA(n_components=8) x_new = pca_new.fit_transform(x) # Splitting the dataset into the Training set and Test set X_train, X_test, y_train, y_test = train_test_split(x_new, y, test_size = 0.2, random_state = 1) def classifier(model): model.fit(X_train,y_train) y_pred=model.predict(X_test) score=accuracy_score(y_pred,y_test) return score*100 classifier(KNeighborsClassifier(n_neighbors=100)),classifier(RandomForestClassifier(n_estimators=100)),classifier(LogisticRegression()),classifier(GaussianNB()) <jupyter_output><empty_output>
no_license
/notebooks/loaiabdalslam/classification-pca-kernel-99.ipynb
Sayem-Mohammad-Imtiaz/kaggle-notebooks
3
<jupyter_start><jupyter_text># Generate content<jupyter_code>wav_out = "/Users/tmiano/Documents/Projects/cvr/me_drilling-sound_171128.wav" img_out = "/Users/tmiano/Documents/Projects/cvr/me_drilling-sound_171128.jpg" sr=44100 duration = 4 # seconds mic_recording = sd.rec(duration * sr, samplerate=sr, channels=2,dtype='float64') print("Recording Audio") sd.wait() print("Audio recording complete , Play Audio") sd.play(mic_recording, sr) sd.wait() print("Play Audio Complete") wav.write(wav_out,sr,mic_recording) print("Audio written.") generate_spectrogram(mic_recording[:,0],sr,img_out)<jupyter_output><empty_output><jupyter_text># Review recordings<jupyter_code>wav_out = "/Users/tmiano/Documents/Projects/cvr/me_barking_171128.wav" sr, myrecording = wav.read(wav_out)<jupyter_output><empty_output>
no_license
/final_project/notebooks/sandbox/mic_audio_spectrogram.ipynb
thommiano/cvr_6501-009
2
<jupyter_start><jupyter_text># WeatherPy ---- ### Analysis * As expected, the weather becomes significantly warmer as one approaches the equator (0 Deg. Latitude). More interestingly, however, is the fact that the southern hemisphere tends to be warmer this time of year than the northern hemisphere. This may be due to the tilt of the earth. * There is no strong relationship between latitude and cloudiness. However, it is interesting to see that a strong band of cities sits at 0, 80, and 100% cloudiness. * There is no strong relationship between latitude and wind speed. However, in northern hemispheres there is a flurry of cities with over 20 mph of wind. --- #### Note * Instructions have been included for each segment. You do not have to follow them exactly, but they are included to help you think through the steps.<jupyter_code># Dependencies and Setup import matplotlib.pyplot as plt import pandas as pd import numpy as np import requests import time from random import uniform from citipy import citipy from datetime import datetime import seaborn as sns import urllib from urllib.error import HTTPError from datetime import datetime # Import API key from api_keys import api_key # Incorporated citipy to determine city based on latitude and longitude from citipy import citipy # Output File (CSV) output_data_file = "output_data/cities.csv" # Range of latitudes and longitudes lat_range = (-90, 90) lng_range = (-180, 180)<jupyter_output><empty_output><jupyter_text>## Generate Cities List<jupyter_code># List for holding lat_lngs and cities lat_lngs = [] cities = [] lat = [] lng = [] cloudiness = [] dt = [] temp = [] humidity = [] wind = [] country = [] # Create a set of random lat and lng combinations lats = np.random.uniform(low=-90.000, high=90.000, size=1500) lngs = np.random.uniform(low=-180.000, high=180.000, size=1500) lat_lngs = zip(lats, lngs) # Identify nearest city for each lat, lng combination for lat_lng in lat_lngs: city = citipy.nearest_city(lat_lng[0], lat_lng[1]).city_name # If the city is unique, then add it to a our cities list if city not in cities: cities.append(city) # Print the city count to confirm sufficient count len(cities)<jupyter_output><empty_output><jupyter_text>### Perform API Calls * Perform a weather check on each city using a series of successive API calls. * Include a print log of each city as it'sbeing processed (with the city number and city name). <jupyter_code>api_key url = "http://api.openweathermap.org/data/2.5/weather?" units = "imperial" # query URL query_url = url + "appid=" + api_key + "&units=" + units + "&q=" print("Beginning Data Retrival") print("------------------------------") i = 1 j = 1 weather_df = {"City":[],"Cloudiness":[],"Country":[],"Date":[],"Humidity":[],"Lat":[],"Lng":[],"Max Temp":[],"Wind Speed": []} for city in cities: response = requests.get(query_url + city) response_json = response.json() if response.status_code == 200: weather_df["City"].append(city) weather_df["Cloudiness"].append(response_json['clouds']['all']) weather_df["Country"].append(response_json['sys']['country']) weather_df["Date"].append(response_json['dt']) weather_df["Humidity"].append(response_json['main']['humidity']) weather_df["Lat"].append(response_json['coord']['lat']) weather_df["Lng"].append(response_json['coord']['lon']) weather_df["Max Temp"].append(response_json['main']['temp_max']) weather_df["Wind Speed"].append(response_json['wind']['speed']) if j <= 50: print(f"Processing Record {j} of Set {i} | {city}") j = j + 1 else: j = 0 i = i + 1 print(f"Processing Record {j} of Set {i} | {city}") j = j + 1 print("City not found. Skipping...") print("-------------------------") print("Data Retrieval Complete") print("-------------------------") <jupyter_output>Beginning Data Retrival ------------------------------ Processing Record 1 of Set 1 | rocha City not found. Skipping... Processing Record 2 of Set 1 | ushuaia City not found. Skipping... Processing Record 3 of Set 1 | cape town City not found. Skipping... Processing Record 4 of Set 1 | lebu City not found. Skipping... Processing Record 5 of Set 1 | hilo City not found. Skipping... Processing Record 6 of Set 1 | punta arenas City not found. Skipping... Processing Record 7 of Set 1 | ocos City not found. Skipping... Processing Record 8 of Set 1 | bluff City not found. Skipping... Processing Record 9 of Set 1 | touros City not found. Skipping... Processing Record 10 of Set 1 | hami City not found. Skipping... Processing Record 11 of Set 1 | tuatapere City not found. Skipping... Processing Record 12 of Set 1 | hasaki City not found. Skipping... Processing Record 13 of Set 1 | talnakh City not found. Skipping... Processing Record 14 of Set 1 | thompson City not found. Skipping... Processing[...]<jupyter_text>### Convert Raw Data to DataFrame * Export the city data into a .csv. * Display the DataFrame<jupyter_code>weather_df = pd.DataFrame(weather_df) weather_df.head() weather_df.count() # convert data to csv weather_df.to_csv('weather_df.csv', encoding='utf-8', index=False) weather_df.shape<jupyter_output><empty_output><jupyter_text>### Plotting the Data * Use proper labeling of the plots using plot titles (including date of analysis) and axes labels. * Save the plotted figures as .pngs.#### Latitude vs. Temperature Plot<jupyter_code>plt.scatter(weather_df["Lat"],weather_df["Max Temp"], marker="o", alpha=0.8, edgecolor="black", color='blue') plt.title("City Latitude vs. Max Temperature (08/22/18)") plt.xlabel("Latitude") plt.ylabel("Max Temperature (F)") plt.ylim(0,104) plt.grid(True) plt.savefig("LatVsTemp.png") plt.show()<jupyter_output><empty_output><jupyter_text>#### Latitude vs. Humidity Plot<jupyter_code>plt.scatter(weather_df["Lat"],weather_df["Humidity"], marker="o", alpha=0.8, edgecolor="black", color='blue') plt.title("City Latitude vs. Humidity (08/22/18)") plt.xlabel("Latitude") plt.ylabel("Humidity (%)") plt.ylim(0,104) plt.grid(True) plt.savefig("LatVsHumidity.png") plt.show()<jupyter_output><empty_output><jupyter_text>#### Latitude vs. Cloudiness Plot<jupyter_code>plt.scatter(weather_df["Lat"],weather_df["Cloudiness"], marker="o", alpha=0.8, edgecolor="black", color='blue') plt.title("City Latitude vs. Cloudiness (08/22/18)") plt.xlabel("Latitude") plt.ylabel("Cloudiness (%)") plt.ylim(0,104) plt.grid(True) plt.savefig("LatVsCloudiness.png") plt.show()<jupyter_output><empty_output><jupyter_text>#### Latitude vs. Wind Speed Plot<jupyter_code>plt.scatter(weather_df["Lat"],weather_df["Wind Speed"], marker="o", alpha=0.8, edgecolor="black", color='blue') plt.title("City Latitude vs. Wind Speed (08/22/18)") plt.xlabel("Latitude") plt.ylabel("Wind Speed (mph)") plt.ylim(0,38) plt.grid(True) plt.savefig("LatVsWindSpeed.png") plt.show()<jupyter_output><empty_output>
no_license
/starter_code/WeatherPy.ipynb
nebajuinpou/Weather_challenge_Python
8
<jupyter_start><jupyter_text>## Dimensionality Reduction### PCA<jupyter_code>from sklearn.decomposition import PCA import numpy as np X = np.random.rand(100, 3) y = 4 + 3 * X.dot(X.T) + np.random.randn(100, 1) pca = PCA(n_components = 2) X2D = pca.fit_transform(X) print(pca.explained_variance_ratio_) pca = PCA() pca.fit(X) cumsum = np.cumsum(pca.explained_variance_ratio_) d = np.argmax(cumsum >= 0.95) + 1 print(d) pca = PCA(n_components=0.95) X_reduced = pca.fit_transform(X) %matplotlib inline import matplotlib import matplotlib.pyplot as plt plt.plot(cumsum) plt.show() from sklearn.datasets import fetch_mldata mnist = fetch_mldata('MNIST original') X, y = mnist["data"], mnist["target"] pca = PCA() pca.fit(X) cumsum = np.cumsum(pca.explained_variance_ratio_) d = np.argmax(cumsum >= 0.95) + 1 plt.plot(cumsum) plt.show() pca = PCA(n_components = 154) X_mnist_reduced = pca.fit_transform(X) X_mnist_recovered = pca.inverse_transform(X_mnist_reduced) # Incremental PCA. from sklearn.decomposition import IncrementalPCA n_batches = 100 inc_pca = IncrementalPCA(n_components=154) for X_batch in np.array_split(X, n_batches): inc_pca.partial_fit(X_batch) X_mnist_reduced = inc_pca.transform(X) # Randomized PCA. rnd_pca = PCA(n_components=154, svd_solver="randomized") X_reduced = rnd_pca.fit_transform(X)<jupyter_output><empty_output><jupyter_text>## Kernel PCA<jupyter_code>from sklearn.decomposition import KernelPCA X = np.random.rand(100, 3) y = 4 + 3 * X.dot(X.T) + np.random.randn(100, 1) rbf_pca = KernelPCA(n_components = 2, kernel="rbf", gamma=0.04) X_reduced = rbf_pca.fit_transform(X) X_mnist, y_mnist = mnist["data"], mnist["target"] shuffled_indices = np.random.permutation(len(X_mnist))[:40] X = np.random.rand(100, 2) y = np.random.rand(100, 1) > 0.5 y.shape from sklearn.model_selection import GridSearchCV from sklearn.linear_model import LogisticRegression from sklearn.pipeline import Pipeline clf = Pipeline([ ("kpca", KernelPCA(n_components=2)), ("log_reg", LogisticRegression()) ]) param_grid = [{ "kpca__gamma": np.linspace(0.03, 0.05, 10), "kpca__kernel": ["rbf", "sigmoid"] }] grid_search = GridSearchCV(clf, param_grid, cv=3) grid_search.fit(X, y) print(grid_search.best_params_)<jupyter_output>{'kpca__gamma': 0.03, 'kpca__kernel': 'rbf'} <jupyter_text>## LLE<jupyter_code>from sklearn.manifold import LocallyLinearEmbedding from sklearn import datasets X, y = datasets.make_swiss_roll() lle = LocallyLinearEmbedding(n_components=2, n_neighbors=10) X_reduced = lle.fit_transform(X)<jupyter_output><empty_output>
no_license
/notebooks/dimensionality-reduction.ipynb
Modest-as/machine-learning-exercises
3
<jupyter_start><jupyter_text># Exercise 02 - OLAP Cubes - Grouping SetsAll the databases table in this demo are based on public database samples and transformations - `Sakila` is a sample database created by `MySql` [Link](https://dev.mysql.com/doc/sakila/en/sakila-structure.html) - The postgresql version of it is called `Pagila` [Link](https://github.com/devrimgunduz/pagila) - The facts and dimension tables design is based on O'Reilly's public dimensional modelling tutorial schema [Link](http://archive.oreilly.com/oreillyschool/courses/dba3/index.html) Start by connecting to the database by running the cells below. If you are coming back to this exercise, then uncomment and run the first cell to recreate the database. If you recently completed the slicing and dicing exercise, then skip to the second cell.<jupyter_code> !PGPASSWORD=student createdb -h 127.0.0.1 -U student pagila_star !PGPASSWORD=student psql -q -h 127.0.0.1 -U student -d pagila_star -f Data/pagila-star.sql<jupyter_output> set_config ------------ (1 row) setval -------- 200 (1 row) setval -------- 605 (1 row) setval -------- 16 (1 row) setval -------- 600 (1 row) setval -------- 109 (1 row) setval -------- 599 (1 row) setval -------- 1 (1 row) setval -------- 1 (1 row) setval -------- 1 (1 row) setval -------- 1 (1 row) setval -------- 16049 (1 row) setval -------- 1000 (1 row) setval -------- 4581 (1 row) setval -------- 6 (1 row) setval -------- 32098 (1 row) setval -------- 16049 (1 row) setval -------- 2 (1 row) setval -------- 2 (1 row) <jupyter_text>### Connect to the local database where Pagila is loaded<jupyter_code>import sql %reload_ext sql DB_ENDPOINT = "127.0.0.1" DB = 'pagila_star' DB_USER = 'student' DB_PASSWORD = 'student' DB_PORT = '5432' # postgresql://username:password@host:port/database conn_string = "postgresql://{}:{}@{}:{}/{}" \ .format(DB_USER, DB_PASSWORD, DB_ENDPOINT, DB_PORT, DB) print(conn_string) %sql $conn_string<jupyter_output>postgresql://student:[email protected]:5432/pagila_star <jupyter_text>### Star Schema# Grouping Sets - It happens often that for 3 dimensions, you want to aggregate a fact: - by nothing (total) - then by the 1st dimension - then by the 2nd - then by the 3rd - then by the 1st and 2nd - then by the 2nd and 3rd - then by the 1st and 3rd - then by the 1st and 2nd and 3rd - Since this is very common, and in all cases, we are iterating through all the fact table anyhow, there is a more clever way to do that using the SQL grouping statement "GROUPING SETS" ## Total Revenue TODO: Write a query that calculates total revenue (sales_amount)<jupyter_code>%%sql SELECT SUM(fs.sales_amount) as revenue FROM factSales fs<jupyter_output> * postgresql://student:***@127.0.0.1:5432/pagila_star 1 rows affected. <jupyter_text>## Revenue by Country TODO: Write a query that calculates total revenue (sales_amount) by country<jupyter_code>%%sql SELECT ds.country, SUM(fs.sales_amount) as revenue FROM factSales fs JOIN dimstore ds ON fs.store_key = ds.store_key GROUP BY ds.country<jupyter_output> * postgresql://student:***@127.0.0.1:5432/pagila_star 2 rows affected. <jupyter_text>## Revenue by Month TODO: Write a query that calculates total revenue (sales_amount) by month<jupyter_code>%%sql SELECT dd.month, SUM(fs.sales_amount) as revenue FROM factSales fs JOIN dimdate dd ON fs.date_key = dd.date_key GROUP BY dd.month ORDER BY dd.month asc, revenue DESC<jupyter_output> * postgresql://student:***@127.0.0.1:5432/pagila_star 5 rows affected. <jupyter_text>## Revenue by Month & Country TODO: Write a query that calculates total revenue (sales_amount) by month and country. Sort the data by month, country, and revenue in descending order. The first few rows of your output should match the table below.<jupyter_code>%%sql SELECT dd.month, ds.country, SUM(fs.sales_amount) as revenue FROM factSales fs JOIN dimdate dd ON fs.date_key = dd.date_key JOIN dimstore ds ON fs.store_key = ds.store_key GROUP BY dd.month, ds.country ORDER BY dd.month asc, ds.country asc, revenue DESC<jupyter_output> * postgresql://student:***@127.0.0.1:5432/pagila_star 10 rows affected. <jupyter_text> month country revenue 1 Australia 2364.19 1 Canada 2460.24 2 Australia 4895.10 2 Canada 4736.78 3 Australia 12060.33 ## Revenue Total, by Month, by Country, by Month & Country All in one shot TODO: Write a query that calculates total revenue at the various grouping levels done above (total, by month, by country, by month & country) all at once using the grouping sets function. Your output should match the table below.<jupyter_code>%%time %%sql SELECT dd.month, ds.country, SUM(fs.sales_amount) as revenue FROM factSales fs JOIN dimdate dd ON fs.date_key = dd.date_key JOIN dimstore ds ON fs.store_key = ds.store_key GROUP BY grouping sets ((), dd.month, ds.country, (dd.month, ds.country))<jupyter_output> * postgresql://student:***@127.0.0.1:5432/pagila_star 18 rows affected. CPU times: user 4.19 ms, sys: 686 µs, total: 4.88 ms Wall time: 34.5 ms
no_license
/course_DE/Udacity-Data-Engineering-master/Cloud Data Warehouse/L1 E2 - Grouping Sets.ipynb
yennanliu/analysis
7
<jupyter_start><jupyter_text>EOS 491/526 Assignment #1 Daniel Scanks V00788200 Question #1<jupyter_code>import numpy as np import matplotlib.pyplot as plt %matplotlib inline n1= np.random.uniform(-1,1,200000) # 1 uniformly distributed random variable on [-1,1] with mean and std dev calculated mean = np.mean(n1) std = np.std(n1) print('mean, standard deviaton') print (mean,std) gaussian = np.random.normal(mean,std,200000) plt.hist(gaussian, bins=50, histtype='step', normed=True, color='r', label='Gaussian') plt.hist(n1, bins=50, histtype='stepfilled', normed=True, color='b', alpha=0.5, label='Uniform Distribution') #histogram plt.title("Gaussian/Uniform Histogram") plt.xlabel("Value") plt.ylabel("Probability") plt.legend() plt.show() n1=np.random.uniform(-1,1,200000) # 2 uniformly distributed variables added w/ same calcuations and plot n2=np.random.uniform(-1,1,200000) n= n1+n2 mean = np.mean(n) std = np.std(n) print('mean, standard deviaton') print (mean,std) gaussian = np.random.normal(mean,std,200000) plt.hist(gaussian, bins=50, histtype='step', normed=True, color='r', label='Gaussian') plt.hist(n, bins=50, histtype='stepfilled', normed=True, color='b', alpha=0.5, label='Uniform Distribution') plt.title("Gaussian/Uniform Histogram") plt.xlabel("Value") plt.ylabel("Probability") plt.legend() plt.show() n1=np.random.uniform(-1,1,200000) # 5 uniformly distributed variables added w/ same calcuations and plot n2=np.random.uniform(-1,1,200000) n3=np.random.uniform(-1,1,200000) n4=np.random.uniform(-1,1,200000) n5=np.random.uniform(-1,1,200000) n= n1+n2+n3+n4+n5 mean = np.mean(n) std = np.std(n) print('mean, standard deviaton') print (mean,std) gaussian = np.random.normal(mean,std,200000) plt.hist(gaussian, bins=50, histtype='step', normed=True, color='r', label='Gaussian') plt.hist(n, bins=50, histtype='stepfilled', normed=True, color='b', alpha=0.5, label='Uniform Distribution') plt.title("Gaussian/Uniform Histogram") plt.xlabel("Value") plt.ylabel("Probability") plt.legend() plt.show() <jupyter_output>mean, standard deviaton (0.0008137093395222234, 0.57696430269689469)
no_license
/.ipynb_checkpoints/EOS491Assignment#1-checkpoint.ipynb
scanks/EOS491Assignments
1
<jupyter_start><jupyter_text> ## [mlcourse.ai](https://mlcourse.ai) – Open Machine Learning Course Authors: [Vitaliy Radchenko](https://www.linkedin.com/in/vitaliyradchenk0/), [Yury Kashnitsky](https://yorko.github.io), and Mikalai Parshutsich. Translated and edited by [Christina Butsko](https://www.linkedin.com/in/christinabutsko/), Artem Gruzdev, [Egor Polusmak](https://www.linkedin.com/in/egor-polusmak/), [Anastasia Manokhina](https://www.linkedin.com/in/anastasiamanokhina/), [Anna Shirshova](http://linkedin.com/in/anna-shirshova-b908458b), and [Yuanyuan Pao](https://www.linkedin.com/in/yuanyuanpao/). This material is subject to the terms and conditions of the [Creative Commons CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) license. Free use is permitted for any non-commercial purpose. *This is a static version of a Jupyter notebook. You can also check out the latest version in the [course repository](https://github.com/Yorko/mlcourse.ai), the corresponding interactive web-based [Kaggle Notebook](https://www.kaggle.com/kashnitsky/topic-5-ensembles-part-3-feature-importance) or a [video lecture](https://www.youtube.com/watch?v=neXJL-AqI_c).*# Topic 5. Ensembles and random forest ## Part 3. Feature importance## Article outline 1. [Intuition](#1.-Intuition) 2. [Illustrating permutation importance](#2.-Illustrating-permutation-importance) 3. [Sklearn Random Forest Feature Importance](#3.-Sklearn-Random-Forest-Feature-Importance) 4. [Practical example](#4.-Practical-example) 5. [Demo assignment](#5.-Demo-assignment) 6. [Useful resources](#6.-Useful-resources) It's quite often that you want to make out the exact reasons of the algorithm outputting a particular answer. Or at the very least to find out which input features contributed most to the result. With Random Forest, you can obtain such information quite easily.## 1. Intuition From the picture below, it is intuitively clear that, in our credit scoring problem, *Age* is much more important than *Income*. This can be formally explained using the concept of *information gain*. In the case of many decision trees or a random forest, the closer the mean position of a feature over all the trees to the root, the more significant it is for a given classification or regression problem. Gains in the splitting criterion, such as the *Gini impurity*, obtained at each optimal split in every tree is a measure of importance that is directly associated with the splitting feature. The value of this score is distinct for each feature and accumulates over all the trees. Let's go a little deeper into the details. There exist a lot of methods to assess feature importances. Leo Breinman in his works suggested to evaluate the importance of a variable by measuring decrease of accuracy of the forest when the variable is randomly permuted or decrease of impurity of a nodes where the given variable is used for splitting. The former method is often called **permutation importance**. The latter method is used in `sklearn`.### Permutation importance Inspired by [this](https://www.researchgate.net/publication/5231126_Conditional_Variable_Importance_for_Random_Forests) article. The average reduction in accuracy caused by a variable is determined during the calculation of the out-of-bag error. The greater the reduction in accuracy due to an exclusion or permutation of the variable, the higher its *importance score*. For this reason, variables with a greater average reduction in accuracy are generally more significant for classification. The rationale for calculating permutation importance is the following: By randomly permuting the predictor variable $X_j$, its original association with the response $Y$ is broken. When the permuted variable $X_j$, together with all the others non-permuted variables, is used the response for the out-of-bag observations, the prediction *accuracy* decreases substantially if the original $X_j$ was associated with response. Thus, as a measure of variable importance, the difference in prediction accuracy before and after permuting is used.More formally: denote $\overline{\mathfrak{B}}^{(t)}$ as the out-of-bag sample for a tree $t$, for $t\in\{1, ..., N\}$ where $N$ is the number of trees in ensemble. Then the permutation importance of variable $X_j$ in tree $t$ is $${PI}^{(t)}\left(X_j\right)=\frac{\sum_{i\in\overline{\mathfrak{B}}^{(t)}}I\left(y_i=\hat{y}_i^{(t)}\right)}{\left|\overline{\mathfrak{B}}^{(t)}\right|}-\frac{\sum_{i\in\overline{\mathfrak{B}}^{(t)}}I\left(y_i=\hat{y}_{i,\pi_j}^{(t)}\right)}{\left|\overline{\mathfrak{B}}^{(t)}\right|}$$ where $\hat{y}_i^{(t)}=f^{(t)}(\mathbf{x}_i)$ is the predicted class for observation $i$ before and $\hat{y}_{i, \pi_j}^{(t)}=f^{(t)}(\mathbf{x}_{i,\pi_j})$ is the predicted class for observation $i$ after permuting $X_j$, $\mathbf{x}_{i,\pi_j}=\left(x_{i,1}, ..., x_{i,j-1},x_{\pi_j(i),j},x_{i,j+1},...,x_{i,p}\right)$ Note that by definition ${PI}^{(t)}=0$, if variable $X_j$ isn't in tree $t$. Now, we can give the feature importance calculation for ensembles: * not normalized: $${PI}\left(X_j\right)=\frac{\sum_{t=1}^N {PI}^{(t)}(X_j)}{N}$$ * normalized by the standard deviation of the differences: $$z_j=\frac{{PI}\left(X_j\right)}{\frac{\hat{\sigma}}{\sqrt{N}}}$$## 2. Illustrating permutation importance Let's assume that we have a toy dataset with 10 instances. Target variable can be either **'N'** or **'P'**. $$\begin{array}{c|c|c|c|c|c|c|c|c|c} \text{Instances}, i & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & \\ \hline y_i & N & P & P & N & N & P & N & N & N & P \\ \end{array}$$ We build an ensemble of 5 trees $t$, for $t\in\{1, ..., 5\}$. For each tree we get out-of-bag sample (denoted $\overline{\mathfrak{B}}^{(t)}$ above). For example for the first tree out-of-bag sample consists of instances # 2, 4, 5, and 6. $$\begin{array}{c|c|c|c|c|c|c|c|c|c} \text{Tree 1} & \text{Bootstrap-sample 1} & 10 & 9 & 7 & 8 & 1 & 3 & 9 & 10 & 10 & 7\\ \hline \text{Tree 2} & \text{Bootstrap-sample 2} & 4 & 8 & 5 & 8 & 3 & 9 & 2 & 6 & 1 & 6\\ \hline \text{Tree 3} & \text{Bootstrap-sample 3} & 6 & 2 & 6 & 10 & 2 & 10 & 3 & 6 & 5 & 1\\ \hline \text{Tree 4} & \text{Bootstrap-sample 4} & 6 & 7 & 8 & 10 & 6 & 10 & 9 & 10 & 8 & 2\\ \hline \text{Tree 5} & \text{Bootstrap-sample 5} & 5 & 8 & 1 & 8 & 5 & 7 & 10 & 1 & 10 & 9\\ \end{array}$$ Thus, out-of-bag samples for each tree $t$ are $$\begin{array}{c|cccc} \text{Tree}, t & \overline{\mathfrak{B}}^{(t)} \\ \hline \text{Tree 1} & 2 & 4 & 5 & 6\\ \hline \text{Tree 2} & 7 & 10\\ \hline \text{Tree 3} & 4 & 7 & 8 & 9\\ \hline \text{Tree 4} & 1 & 3 & 4 & 5\\ \hline \text{Tree 5} & 2 & 3 & 4 & 6\\ \hline \end{array}$$Suppose that we have four features $X_j$, $j\in\{1, 2, 3, 4\}$ and we'd like to compute _permutation importance_ for $X_2$. First, for each out-of-bag sample we compute _accuracy_ of the model before and after permutation of the values of $X_2$. For instance, before permutation for $\overline{\mathfrak{B}}^{(1)}$ we have $$\begin{array}{c|cccc|cc|c} & X_1 & \color{red}{X_2} & X_3 & X_4 & y_i & \hat{y}_i & I\left(y_i=\hat{y}_i\right)\\ \hline \textbf{2} & 1 & \color{red}2 & 11 & 101 & \textbf{P} & \textbf{P} & 1\\ \hline \textbf{4} & 2 & \color{red}3 & 12 & 102 & \textbf{N} & \textbf{P} & 0\\ \hline \textbf{5} & 3 & \color{red}5 & 13 & 103 & \textbf{N} & \textbf{N} & 1\\ \hline \textbf{6} & 4 & \color{red}7 & 14 & 104 & \textbf{P} & \textbf{P} & 1\\ \end{array}$$ Thus, the accuracy before permutation is $3/4=0.75$. After permutation for $\overline{\mathfrak{B}}^{(1)}$ we have $$\begin{array}{c|cccc|cc|c} & X_1 & \color{red}{X_2} & X_3 & X_4 & y_i & \hat{y}_i & I\left(y_i=\hat{y}_i\right)\\ \hline \textbf{2} & 1 & \color{red}5 & 11 & 101 & \textbf{P} & \textbf{P} & 0\\ \hline \textbf{4} & 2 & \color{red}7 & 12 & 102 & \textbf{N} & \textbf{P} & 0\\ \hline \textbf{5} & 3 & \color{red}2 & 13 & 103 & \textbf{N} & \textbf{N} & 1\\ \hline \textbf{6} & 4 & \color{red}3 & 14 & 104 & \textbf{P} & \textbf{P} & 1\\ \end{array}$$ The accuracy after permutation is $2/4=0.50$. Then the difference between accuracies is computed. The above mentioned steps are to be done for each out-of-bag sample $\overline{\mathfrak{B}}^{(t)}$. To get not normalized _permutation importance_ we sum all computed differences and divide by the number of trees. Normalization is done by dividing _not normalized permutation importance_ by standard error.## 3. Sklearn Random Forest Feature Importance Inspired by [this](https://medium.com/@srnghn/the-mathematics-of-decision-trees-random-forest-and-feature-importance-in-scikit-learn-and-spark-f2861df67e3) article. Sklearn library uses another approach to determine feature importance. The rationale for that method is that the more gain in information the node (with splitting feature $X_j$) provides, the higher its importance. The average reduction in the Gini impurity – or MSE for regression – represents the contribution of each feature to the homogeneity of nodes and leaves in the resulting Random Forest model. Each time a selected feature is used for splitting, the Gini impurity of the child nodes is calculated and compared with that of the original node. Gini impurity is a score of homogeneity with the range from 0 (homogeneous) to 1 (heterogeneous). The changes in the value of the splitting criterion are accumulated for each feature and normalized at the end of the calculation. A higher reduction in the Gini impurity signals that splitting results by this feature results in nodes with higher purity. The algorithm of obtaining feature importance may be represented with the following sequence of steps: 1. For each tree $t$ in ensemble $t\in\{1,...,N\}$: 1.1. for each node $i$ calculate the reduction in impurity (or MSE, or entropy) as ${RI}_i^{(t)}=w_i^{(t)}\cdot I_i^{(t)} - w_{LEFT_i}^{(t)}\cdot I_{LEFT_i}^{(t)}-w_{RIGHT_i}^{(t)}\cdot I_{RIGHT_i}^{(t)}$, where: - $w_i^{(t)}$, $w_{LEFT_i}^{(t)}$, and $w_{RIGHT_i}^{(t)}$ are respectively weighted number of samples reaching node $i$ in tree $t$, and its left $LEFT_i$ and right $RIGHT_i$ children - $I_i^{(t)}$, $I_{LEFT_i}^{(t)}$, $I_{RIGHT_i}^{(t)}$ are impurity of the nodes. For leaves ${RI}_i^{(t)}$ is equal to 0. 1.2. for each feature $j$ calculate its importance in that particular tree as $${FI}_j^{(t)}=\frac{\sum_{i:\text{node }i\text{ splits on feature } j}{RI}_i^{(t)}}{\sum_{i\in\text{all nodes}}{RI}_i^{(t)}}$$ That means that in numerator we sum the reduction in impurity only in those nodes where feature $j$ is situated. 2. Calculate the average feature importances over all trees in ensemble: $${FI}_j=\frac{\sum_{t=1}^N {FI}_j^{(t)}}{N}$$Those are pretty confusing formulas so let's demonstrate each step with the Iris Dataset.<jupyter_code>import numpy as np import pandas as pd from sklearn.datasets import load_iris import matplotlib.pyplot as plt import seaborn as sns iris = load_iris() data = iris['data'] target = iris['target'] data = pd.DataFrame(data, columns=iris['feature_names']) data.head()<jupyter_output><empty_output><jupyter_text>Since our aim is just to demonstrate the sequence of steps in calculating feature importances we'll transform the `target` variable as for classifying Iris Virginica One-To-All.<jupyter_code>target = pd.Series(target).map({0: 0, 1: 0, 2: 1})<jupyter_output><empty_output><jupyter_text>Creating Random Forest. For reproducibility, we set `random_state=17`. For the sake of simplicity we set the number of trees to 3 and limit the depth of trees in ensemble to be not greater than 3.<jupyter_code>from sklearn.ensemble import RandomForestClassifier rfc = RandomForestClassifier(n_estimators=3, max_depth=3, random_state=17) rfc.fit(data, target);<jupyter_output><empty_output><jupyter_text>After fitting list of all the trees are stored in `estimators_` property.<jupyter_code>tree_list = rfc.estimators_<jupyter_output><empty_output><jupyter_text>Visualizing trees<jupyter_code>from sklearn import tree plt.figure(figsize=(16,12)) tree.plot_tree(tree_list[0], filled=True, feature_names=iris['feature_names'], class_names=['Y', 'N'], node_ids=True); plt.figure(figsize=(16,12)) tree.plot_tree(tree_list[1], filled=True, feature_names=iris['feature_names'], class_names=['Y', 'N'], node_ids=True); plt.figure(figsize=(6,4)) tree.plot_tree(tree_list[2], filled=True, feature_names=iris['feature_names'], class_names=['Y', 'N'], node_ids=True);<jupyter_output><empty_output><jupyter_text>Let's start from the first tree and `Sepal length (cm)` feature. This feature is located in two nodes: the root (#0) and the rightmost node (#4). The reduction in impurity for these nodes are: $${RI}_{{SL}_1}^{(1)}=\frac{150}{150}\cdot 0.482578 - \frac{63}{150}\cdot 0.061476 - \frac{87}{150}\cdot 0.436517 = 0.203578$$ $${RI}_{{SL}_2}^{(1)}=\frac{56}{150}\cdot 0.035077 - \frac{7}{150}\cdot 0.244898 - \frac{49}{150}\cdot 0 = 0.001667$$ Note: The impurity for each node was recalculated to gain more accuracy than given in the picture. By doing the same calculations we get the following reduction in impurity for `Petal width (cm)`, and `Petal width (cm)` features: $${RI}_{PL}^{(1)}=0.035785$$ $${RI}_{{PW}_1}^{(1)}=0.025820$$ $${RI}_{{PW}_2}^{(1)}=0.193633$$ Summarizing all numbers in table $$\begin{array}{c|cc} \text{Feature}, j & \text{Total }RI_j^{(1)} & {FI}_j^{(1)}\\ \hline SL & 0.205244 & 0.445716\\ SW & 0.000000 & 0.000000\\ PL & 0.035785 & 0.077712\\ PW & 0.219453 & 0.476572\\ \hline \sum & 0.460483 \end{array}$$ After performing the same calculations for the second and third tree we average the results for features: $$\begin{array}{c|ccc|c} \text{Feature}, j & {FI}_j^{(1)}& {FI}_j^{(2)}& {FI}_j^{(3)} & {FI}_j\\ \hline SL & 0.445716 & 0.000000 & 0.000000 & 0.148572\\ SW & 0.000000 & 0.039738 & 0.000000 & 0.013246\\ PL & 0.077712 & 0.844925 & 0.162016 & 0.361551\\ PW & 0.476572 & 0.115337 & 0.837984 & 0.476631\\ \end{array}$$Let's compare our result with those stored in the `feature_importances_` attribute.<jupyter_code>print(iris['feature_names']) print(rfc.feature_importances_)<jupyter_output>['sepal length (cm)', 'sepal width (cm)', 'petal length (cm)', 'petal width (cm)'] [0.14857187 0.01324612 0.36155096 0.47663104] <jupyter_text>Voila!## 4. Practical example Let's consider the results of a survey given to visitors of hostels listed on Booking.com and TripAdvisor.com. Our features here are the average ratings for different categories including service quality, room condition, value for money, etc. Our target variable is the hostel's overall rating on the website.<jupyter_code>from sklearn.ensemble.forest import RandomForestRegressor hostel_data = pd.read_csv("../../data/hostel_factors.csv") features = {"f1":u"Staff", "f2":u"Hostel booking", "f3":u"Check-in and check-out", "f4":u"Room condition", "f5":u"Shared kitchen condition", "f6":u"Shared space condition", "f7":u"Extra services", "f8":u"General conditions & conveniences", "f9":u"Value for money", "f10":u"Customer Co-creation"} forest = RandomForestRegressor(n_estimators=1000, max_features=10, random_state=0) forest.fit(hostel_data.drop(['hostel', 'rating'], axis=1), hostel_data['rating']) importances = forest.feature_importances_ indices = np.argsort(importances)[::-1] # Plot the feature importancies of the forest num_to_plot = 10 feature_indices = [ind+1 for ind in indices[:num_to_plot]] # Print the feature ranking print("Feature ranking:") for f in range(num_to_plot): print("%d. %s %f " % (f + 1, features["f"+str(feature_indices[f])], importances[indices[f]])) plt.figure(figsize=(15,5)) plt.title(u"Feature Importance") bars = plt.bar(range(num_to_plot), importances[indices[:num_to_plot]], color=([str(i/float(num_to_plot+1)) for i in range(num_to_plot)]), align="center") ticks = plt.xticks(range(num_to_plot), feature_indices) plt.xlim([-1, num_to_plot]) plt.legend(bars, [u''.join(features["f"+str(i)]) for i in feature_indices]);<jupyter_output>Feature ranking: 1. Staff 0.182757 2. Value for money 0.148373 3. Shared space condition 0.128296 4. Extra services 0.116604 5. Customer Co-creation 0.106668 6. General conditions & conveniences 0.088589 7. Shared kitchen condition 0.074273 8. Check-in and check-out 0.061521 9. Hostel booking 0.053615 10. Room condition 0.039305
no_license
/jupyter_notebooks/Lecture_05_Ensembles/topic-5-ensembles-part-3-feature-importance.ipynb
RisnyK/mlcourse_dubai
7
<jupyter_start><jupyter_text># Churn Prediction For Bank CustomerWe have a dataset in which there are details of a bank's customers and the target variable is a binary variable reflecting the fact whether the customer left the bank (closed his account) or he continues to be a customer.## Dataset - **RowNumber:** corresponds to the record (row) number and has no effect on the output. - **CustomerId:** contains random values and has no effect on customer leaving the bank. - **Surname:** the surname of a customer has no impact on their decision to leave the bank. - **CreditScore:** can have an effect on customer churn, since a customer with a higher credit score is less likely to leave the bank. - **Geography:** a customer’s location can affect their decision to leave the bank. - **Gender:** it’s interesting to explore whether gender plays a role in a customer leaving the bank. - **Age:** this is certainly relevant, since older customers are less likely to leave their bank than younger ones. - **Tenure:** refers to the number of years that the customer has been a client of the bank. Normally, older clients are more loyal and less likely to leave a bank. - **Balance:** also a very good indicator of customer churn, as people with a higher balance in their accounts are less likely to leave the bank compared to those with lower balances. - **NumOfProducts:** refers to the number of products that a customer has purchased through the bank. - **HasCrCard:** denotes whether or not a customer has a credit card. This column is also relevant, since people with a credit card are less likely to leave the bank. - **IsActiveMember:** active customers are less likely to leave the bank. - **EstimatedSalary:** as with balance, people with lower salaries are more likely to leave the bank compared to those with higher salaries. - **Exited:** whether or not the customer left the bank. (0=No,1=Yes)## Improt Libraries<jupyter_code>import numpy as np import pandas as pd import seaborn as sns import matplotlib.pyplot as plt %matplotlib inline # %matplotlib notebook plt.rcParams["figure.figsize"] = (9,6) # plt.rcParams['figure.dpi'] = 100 sns.set_style("whitegrid") import warnings warnings.filterwarnings("ignore") warnings.warn("this will not show") pd.set_option('display.float_format', lambda x: '%.3f' % x)<jupyter_output><empty_output><jupyter_text>## Indest Data<jupyter_code>df = pd.read_csv("Churn_Modelling.csv") df.shape<jupyter_output><empty_output><jupyter_text>The datset has **1000** rows with **14** attributes<jupyter_code>df.head() df.info()<jupyter_output><class 'pandas.core.frame.DataFrame'> RangeIndex: 10000 entries, 0 to 9999 Data columns (total 14 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 RowNumber 10000 non-null int64 1 CustomerId 10000 non-null int64 2 Surname 10000 non-null object 3 CreditScore 10000 non-null int64 4 Geography 10000 non-null object 5 Gender 10000 non-null object 6 Age 10000 non-null int64 7 Tenure 10000 non-null int64 8 Balance 10000 non-null float64 9 NumOfProducts 10000 non-null int64 10 HasCrCard 10000 non-null int64 11 IsActiveMember 10000 non-null int64 12 EstimatedSalary 10000 non-null float64 13 Exited 10000 non-null int64 dtypes: float64(2), int64(9), object(3) memory usage: 1.1+ MB <jupyter_text>## Exploratory Data Analysis and Visualization 1. Implement basic steps to see how is your data looks like 2. Check for missing values 3. Drop the features that not suitable for modelling 4. Implement basic visualization steps such as histogram, countplot, heatmap 5. Convert categorical variables to dummy variables👇👇👇We dropped follow columns. Because these columnns has no effect on customer leaving the bank.👇👇👇<jupyter_code>df = df.drop(["RowNumber","CustomerId","Surname"], axis=1) print(df.isnull().sum().any()) print(df.duplicated().sum())<jupyter_output>False 0 <jupyter_text>👆👆👆We have no miissing value and duplicated value 👆👆👆<jupyter_code>df.describe().T<jupyter_output><empty_output><jupyter_text>## Exited Column<jupyter_code>df["Exited"].value_counts()<jupyter_output><empty_output><jupyter_text>- Our dataset is imbalanced. - Nonexited customers more than exited customers. - Our goal should be identify exited customers accurately<jupyter_code>sns.countplot(df["Exited"]); col = ["Geography","Gender","Tenure","NumOfProducts","HasCrCard","IsActiveMember"] for i in col: print(i.upper()) print(df[i].value_counts()) print("---"*10) fig, ax = plt.subplots(3, 2, figsize=(16, 12)) sns.countplot(x='Geography', hue = 'Exited',data = df, ax=ax[0][0]) sns.countplot(x='Gender', hue = 'Exited',data = df, ax=ax[0][1]) sns.countplot(x='HasCrCard', hue = 'Exited',data = df, ax=ax[1][0]) sns.countplot(x='IsActiveMember', hue = 'Exited',data = df, ax=ax[1][1]); sns.countplot(x='NumOfProducts', hue = 'Exited',data = df, ax=ax[2][0]) sns.countplot(x='Tenure', hue = 'Exited',data = df, ax=ax[2][1]);<jupyter_output><empty_output><jupyter_text> 👆👆👆 - Most of the customers are from France. However, the number of customers churned is less in France and more in Germany compared to the population. - Male customers more than female customers, but more female customers churned - The majority of the customers that churned are those with credit cards - The majority of the customers that churned are those inactive memebers<jupyter_code>fig, ax = plt.subplots(3, 2, figsize=(16, 10)) sns.boxplot(y='CreditScore',x = 'Exited', hue = 'Exited',data = df, ax=ax[0][0]) sns.boxplot(y='Age',x = 'Exited', hue = 'Exited',data = df , ax=ax[0][1]) sns.boxplot(y='Tenure',x = 'Exited', hue = 'Exited',data = df, ax=ax[1][0]) sns.boxplot(y='Balance',x = 'Exited', hue = 'Exited',data = df, ax=ax[1][1]) sns.boxplot(y='NumOfProducts',x = 'Exited', hue = 'Exited',data = df, ax=ax[2][0]) sns.boxplot(y='EstimatedSalary',x = 'Exited', hue = 'Exited',data = df, ax=ax[2][1]);<jupyter_output><empty_output><jupyter_text> 👆👆👆 - There is no significant difference in the credit score distribution between retained and churned customers. - The older customers are churning at more than the younger ones - Customers with high balance scores leave the bank - Number of products and salary amount do not have a significant effect on leaving the bank<jupyter_code>fig, ax = plt.subplots(3, 2, figsize=(16, 12)) sns.countplot(x='Gender', hue = 'Gender',data = df, ax=ax[0][0]) sns.countplot(x='Geography', hue = 'Gender',data = df, ax=ax[0][1]) sns.countplot(x='HasCrCard', hue = 'Gender',data = df, ax=ax[1][0]) sns.countplot(x='IsActiveMember', hue = 'Gender',data = df, ax=ax[1][1]); sns.countplot(x='NumOfProducts', hue = 'Gender',data = df, ax=ax[2][0]) sns.countplot(x='Tenure', hue = 'Gender',data = df, ax=ax[2][1]);<jupyter_output><empty_output><jupyter_text>👆👆👆In all categories shown in the chart, the number of male customers is more than the number of female customers.👆👆👆## Age Column<jupyter_code>plt.figure(figsize = (16, 6)) sns.countplot(x="Age", hue="Exited", data=df);<jupyter_output><empty_output><jupyter_text>👆👆👆The age distribution of active customers and exited customers is almost similar.👆👆👆<jupyter_code>plt.figure(figsize = (12, 8)) sns.histplot(x="CreditScore", data=df,kde = True, hue = "Exited"); plt.figure(figsize = (12, 6)) sns.histplot(x="Balance", data=df, hue = "Exited", kde = True); plt.figure(figsize=(10, 10)) sns.heatmap(df.corr(), annot=True); plt.figure(figsize = (10,6)) df.corr()['Exited'].sort_values().drop("Exited").plot(kind = "barh");<jupyter_output><empty_output><jupyter_text>## Preprocessing of Data - Train | Test Split, Scalling<jupyter_code>df.info() df = pd.get_dummies(df, drop_first=True) from sklearn.model_selection import train_test_split from sklearn.preprocessing import MinMaxScaler X = df.drop('Exited', axis=1) y = df['Exited'] X_train, X_test, y_train, y_test = train_test_split(X, y, stratify = y, test_size = 0.20, random_state = 42) scaler = MinMaxScaler() X_train= scaler.fit_transform(X_train) X_test = scaler.transform(X_test) X_train.shape X_test.shape<jupyter_output><empty_output><jupyter_text>## Modelling & Model Performance### Import related libraries<jupyter_code>from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, Activation from tensorflow.keras.metrics import Accuracy, Recall,Precision from sklearn.metrics import classification_report, confusion_matrix from sklearn.metrics import plot_roc_curve, roc_auc_score, roc_curve from sklearn.model_selection import cross_val_score, cross_validate from tensorflow.keras.optimizers import Adam, SGD,RMSprop from sklearn.model_selection import GridSearchCV from tensorflow.keras.callbacks import EarlyStopping<jupyter_output><empty_output><jupyter_text>### Creating Model### without class_weigth<jupyter_code>model = Sequential() model.add(Dense(44, activation = "relu")) model.add(Dense(22, activation = "relu")) model.add(Dense(11, activation = "relu")) model.add(Dense(1, activation = "sigmoid")) opt = Adam(lr = 0.005) model.compile(optimizer = opt, loss = "binary_crossentropy", metrics = ["Recall"]) early_stop = EarlyStopping(monitor = "recall", mode = "auto", verbose = 1, patience = 15, restore_best_weights=True) model.fit(x = X_train, y = y_train, validation_split = 0.1, batch_size = 32, epochs = 1000, verbose = 1, callbacks = [early_stop])<jupyter_output>Epoch 1/1000 225/225 [==============================] - 1s 2ms/step - loss: 0.4854 - recall: 0.0650 - val_loss: 0.4291 - val_recall: 0.2500 Epoch 2/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.4201 - recall: 0.2625 - val_loss: 0.3857 - val_recall: 0.2039 Epoch 3/1000 225/225 [==============================] - 0s 1ms/step - loss: 0.3897 - recall: 0.3329 - val_loss: 0.3513 - val_recall: 0.3355 Epoch 4/1000 225/225 [==============================] - 0s 1ms/step - loss: 0.3691 - recall: 0.3755 - val_loss: 0.3493 - val_recall: 0.3092 Epoch 5/1000 225/225 [==============================] - 0s 1ms/step - loss: 0.3644 - recall: 0.3829 - val_loss: 0.3422 - val_recall: 0.4145 Epoch 6/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.3610 - recall: 0.3972 - val_loss: 0.3354 - val_recall: 0.3882 Epoch 7/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.3571 - recall: 0.3890 - val_loss: 0.3311 - val_recall: 0.3684 Epoch 8/1000 225/225[...]<jupyter_text>#### Evaluate<jupyter_code>loss_df = pd.DataFrame(model.history.history) loss_df.plot(); y_pred = (model.predict(X_test) > 0.5).astype("int32") #y_pred = model.predict_classes(X_test) print(confusion_matrix(y_test, y_pred)) print(classification_report(y_test, y_pred))<jupyter_output>[[1519 74] [ 314 93]] precision recall f1-score support 0 0.83 0.95 0.89 1593 1 0.56 0.23 0.32 407 accuracy 0.81 2000 macro avg 0.69 0.59 0.61 2000 weighted avg 0.77 0.81 0.77 2000 <jupyter_text>👆👆👆Our results are good, but we need to increase the recall value for 1 values.👆👆👆 👇👇👇For this we will use the class weight parameter in ANN👇👇👇### with class_weigth Investigate how the "class_weight" hyper-parameter is used in a Neural Network.<jupyter_code>from sklearn.utils import class_weight class_weights = dict(zip(np.unique(y_train), class_weight.compute_class_weight('balanced', np.unique(y_train), y_train))) model = Sequential() model.add(Dense(44, activation = "relu")) model.add(Dense(22, activation = "relu")) model.add(Dense(11, activation = "relu")) model.add(Dense(1, activation = "sigmoid")) opt = Adam(lr = 0.005) model.compile(optimizer = opt, loss = "binary_crossentropy", metrics = ["Recall"]) early_stop = EarlyStopping(monitor = "recall", mode = "auto", verbose = 1, patience = 15) model.fit(x = X_train, y = y_train, validation_split = 0.1, batch_size = 32, epochs = 1000, verbose = 1, class_weight = class_weights, callbacks = [early_stop])<jupyter_output>Epoch 1/1000 225/225 [==============================] - 1s 2ms/step - loss: 0.6193 - recall: 0.6590 - val_loss: 0.5079 - val_recall: 0.5724 Epoch 2/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.5450 - recall: 0.7300 - val_loss: 0.4622 - val_recall: 0.7632 Epoch 3/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.5046 - recall: 0.7300 - val_loss: 0.4746 - val_recall: 0.7566 Epoch 4/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.4980 - recall: 0.7429 - val_loss: 0.4798 - val_recall: 0.7368 Epoch 5/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.4887 - recall: 0.7375 - val_loss: 0.4234 - val_recall: 0.7368 Epoch 6/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.4806 - recall: 0.7463 - val_loss: 0.5407 - val_recall: 0.8487 Epoch 7/1000 225/225 [==============================] - 0s 2ms/step - loss: 0.4785 - recall: 0.7368 - val_loss: 0.4143 - val_recall: 0.7237 Epoch 8/1000 225/225[...]<jupyter_text>#### Evaluate<jupyter_code>loss_df = pd.DataFrame(model.history.history) loss_df.plot(); y_pred = (model.predict(X_test) > 0.5).astype("int32") #y_pred = model.predict_classes(X_test) print(confusion_matrix(y_test, y_pred)) print(classification_report(y_test, y_pred))<jupyter_output>[[1278 315] [ 109 298]] precision recall f1-score support 0 0.92 0.80 0.86 1593 1 0.49 0.73 0.58 407 accuracy 0.79 2000 macro avg 0.70 0.77 0.72 2000 weighted avg 0.83 0.79 0.80 2000 <jupyter_text>👆👆👆Accuracy score dropped after class-weight operation, but reacll score increased considerably👆👆👆## GridSearchCV<jupyter_code>from tensorflow.keras.wrappers.scikit_learn import KerasClassifier def build_classifier(optimizer): classifier = Sequential() classifier.add(Dense(units = 44, activation = 'relu')) classifier.add(Dense(units = 22, activation = 'relu')) classifier.add(Dense(units = 11, activation = 'relu')) classifier.add(Dense(units = 1, activation = 'sigmoid')) classifier.compile(optimizer = optimizer, loss = 'binary_crossentropy', metrics = ['Recall']) return classifier early_stop = EarlyStopping(monitor = "recall", mode = "auto", verbose = 1, patience = 15) classifier = KerasClassifier(build_fn = build_classifier, epochs = 200) parameters = {'batch_size': [32, 64], 'optimizer': ['adam', 'rmsprop', "SGD", "adagrad", "adadelta"]} grid_model = GridSearchCV(estimator = classifier, param_grid = parameters, scoring = 'recall', cv = 10, n_jobs = -1, verbose = 1,) grid_model.fit(X_train, y_train, callbacks = [early_stop], class_weight = class_weights) grid_model.best_score_ grid_model.best_params_ y_pred = (grid_model.predict(X_test) > 0.5).astype("int32") #y_pred = model.predict_classes(X_test) print(confusion_matrix(y_test, y_pred)) print(classification_report(y_test, y_pred))<jupyter_output>[[1381 212] [ 147 260]] precision recall f1-score support 0 0.90 0.87 0.88 1593 1 0.55 0.64 0.59 407 accuracy 0.82 2000 macro avg 0.73 0.75 0.74 2000 weighted avg 0.83 0.82 0.83 2000 <jupyter_text>#### Evaluate#### for keras models<jupyter_code>y_pred_proba = model.predict(X_test) fpr, tpr, thresholds = roc_curve(y_test, y_pred_proba) plt.plot([0,1],[0,1],'k--') plt.plot(fpr,tpr, label='ANN') plt.xlabel('fpr') plt.ylabel('tpr') plt.title('ROC curve') plt.show()<jupyter_output><empty_output><jupyter_text>#### for gridsearchcv model<jupyter_code>y_pred_proba = grid_model.predict_proba(X_test)[:,1] fpr, tpr, thresholds = roc_curve(y_test, y_pred_proba) plt.plot([0,1],[0,1],'k--') plt.plot(fpr,tpr, label='ANN') plt.xlabel('fpr') plt.ylabel('tpr') plt.title('ROC curve') plt.show() roc_auc_score(y_test, y_pred_proba)<jupyter_output><empty_output><jupyter_text>## Final Model and Model DeploymentWe use the best parameters that we found with Grid Search in the final model.<jupyter_code>import pickle pickle.dump(scaler, open("scaler_bank", 'wb')) class_weights = dict(zip(np.unique(y_train), class_weight.compute_class_weight('balanced', np.unique(y_train), y_train))) model = Sequential() model.add(Dense(44, activation = "relu")) model.add(Dense(22, activation = "relu")) model.add(Dense(11, activation = "relu")) model.add(Dense(1, activation = "sigmoid")) opt = RMSprop(lr = 0.005) model.compile(optimizer = opt, loss = "binary_crossentropy", metrics = ["Recall"]) early_stop = EarlyStopping(monitor = "recall", mode = "auto", verbose = 1, patience = 15) model.fit(x = X_train, y = y_train, validation_data = (X_test, y_test), batch_size = 32, epochs = 1000, verbose = 1, class_weight = class_weights, callbacks = [early_stop]) loss_df = pd.DataFrame(model.history.history) loss_df.plot() y_pred = (model.predict(X_test) > 0.5).astype("int32") #y_pred = model.predict_classes(X_test) print(confusion_matrix(y_test, y_pred)) print(classification_report(y_test, y_pred))<jupyter_output>[[1212 381] [ 91 316]] precision recall f1-score support 0 0.93 0.76 0.84 1593 1 0.45 0.78 0.57 407 accuracy 0.76 2000 macro avg 0.69 0.77 0.70 2000 weighted avg 0.83 0.76 0.78 2000 <jupyter_text>👆👆👆At this stage, Adam and RMSprop were tested. Learning rate changed. The highest scoring result has been found👆👆👆<jupyter_code>model.save('model_bank.h5')<jupyter_output><empty_output><jupyter_text>### Prediction<jupyter_code>from tensorflow.keras.models import load_model model_bank = load_model('model_bank.h5') scaler_bank = pickle.load(open("scaler_bank", "rb")) customer = df.drop('Exited', axis = 1).iloc[5:10, :] customer customer = scaler_bank.transform(customer) customer (model_bank.predict(customer) > 0.5).astype("int32") df["Exited"].iloc[5:10]<jupyter_output><empty_output><jupyter_text>Our model made an accurate prediction of 80% on the given sample values👏👏👏## Comparison with ML### Logistic Regression<jupyter_code>from sklearn.linear_model import LogisticRegression log_model=LogisticRegression(class_weight="balanced") log_model.fit(X_train, y_train) y_pred = log_model.predict(X_test) print(confusion_matrix(y_test, y_pred)) print(classification_report(y_test, y_pred))<jupyter_output>[[1137 456] [ 122 285]] precision recall f1-score support 0 0.90 0.71 0.80 1593 1 0.38 0.70 0.50 407 accuracy 0.71 2000 macro avg 0.64 0.71 0.65 2000 weighted avg 0.80 0.71 0.74 2000 <jupyter_text>### Random Forest<jupyter_code>from sklearn.ensemble import RandomForestClassifier rf_model = RandomForestClassifier(class_weight="balanced") rf_model.fit(X_train, y_train) y_pred = rf_model.predict(X_test) print(confusion_matrix(y_test, y_pred)) print(classification_report(y_test, y_pred))<jupyter_output>[[1549 44] [ 226 181]] precision recall f1-score support 0 0.87 0.97 0.92 1593 1 0.80 0.44 0.57 407 accuracy 0.86 2000 macro avg 0.84 0.71 0.75 2000 weighted avg 0.86 0.86 0.85 2000
no_license
/Classification_Churn_Modelling.ipynb
bunyamin-polat/Churn-Prediction-with-DL-For-Bank-Customer
25
<jupyter_start><jupyter_text>## Note on spells If child runs away for longer than a week, their spell will be considered ended, and a new spell would start if they return.<jupyter_code># age of child at start of spell ages = spells.STARTAGE.value_counts() _ = plt.bar(ages.index, ages) # primary placement type of spell _ = sns.countplot(spells.TYPE)<jupyter_output><empty_output><jupyter_text>- AL - Alternative Placement (equiv to PAL) - FC - Foster Care (equiv to PFC) - GH Group Home (equiv to PGH) - IL Independent Living (equiv to PIL) - KC Relative/Kinship Care (including fictive kin) – licensed & unlicensed (equiv to PKC) - RC Residential Care (excluding RTF) (equiv to PRC) - RT Residential Treatment Facilities (equiv to PRT) - SF Shelter Foster Care (equiv to PSF) - SG Shelter Group Home (equiv to PSG) - SK Shelter Kinship Care (equiv to PSK) - UK Unknown (equiv to PUK) - MX - Mixed type (When >50% is not equal to particular type)<jupyter_code># number of placements per spell nplacements = spells.NPLACES.value_counts() _ = plt.bar(nplacements.index, nplacements) # CDF of spell durations in days spell_len = lv.get_cdf(spells.DURAT) _ = plt.semilogx(spell_len.index, spell_len) # spell durations vs number of placements _ = sns.jointplot(spells.DURAT, spells.NPLACES, marker='.', alpha = 0.3) # spell exit types _ = sns.countplot(spells.EXIT)<jupyter_output><empty_output><jupyter_text>## EXIT codes in spell_clean.csv ### Good: - XRF: reunified with family - XLC: legal custodianship - XRL: to relative (obsolete code) - XCA: adopted ### Bad: - XOT: other - XOP: other (obsolete code) - XRM: age out - XRY: runaway - XJP: juvenile probation ### Unknown: - ZTC: spell ongoing - XUK: unknown<jupyter_code>#map EXITS to outcome: GOOD, BAD, or UNKNOWN good_val = 'GOOD' bad_val = 'BAD' unknown_val = 'UNKNOWN' exit_mapping = { 'XRF':good_val, 'XLC':good_val, 'XRL':good_val, 'XCA':good_val, 'XOT':bad_val, 'XOP':bad_val, 'XRM':bad_val, 'XRY':bad_val, 'XJP':bad_val, 'ZTC':unknown_val, 'XUK':unknown_val } outcomes = [exit_mapping[x] for x in spells.EXIT if x in exit_mapping] sns.countplot(outcomes) spells['outcome'] = outcomes<jupyter_output><empty_output><jupyter_text># distributions by outcome<jupyter_code>def pmf_by_outcome(col, data=spells, figsize=(12,5), color = sns.color_palette('muted')): fig, ax = plt.subplots(1,1, figsize=figsize) n_total = len(data[col]) # color = ['#77298A', '#94B02A', '#B95B13'] for i,x in enumerate(['UNKNOWN', 'GOOD', 'BAD']): series = data[data.outcome==x][col] pct_string = ' ({0:.1f}%)'.format(100*len(series)/n_total) pmf = lv.get_pmf(series) is_categorical = pmf.index.dtype=='O' xs = np.arange(len(pmf.index))+i/3 if is_categorical else pmf.index+i/3 ax.bar(xs, pmf, 1/3, facecolor=color[i], label=x+pct_string) if is_categorical: ax.set_xticks(range(len(pmf))) else: ax.set_xticks(pmf.index) ax.set_xticklabels(pmf.index, ha='left') ax.legend() def cdf_by_outcome(col, data=spells, figsize=(12,5), scale='lin', color = sns.color_palette('muted') ): fig, ax = plt.subplots(1,1, figsize=figsize) n_total = len(data[col]) for i,x in enumerate([ 'UNKNOWN', 'GOOD', 'BAD']): series = data[data.outcome==x][col] pct_string = ' ({0:.1f}%)'.format(100*len(series)/n_total) cdf = lv.get_cdf(series) plot_fn = ax.plot if scale=='lin' else ax.semilogx plot_fn(cmf.index, cdf, color=color[i], label=x+pct_string) ax.legend() # PMF of number of placements in spell, separated by outcome pmf_by_outcome('NPLACES') cdf_by_outcome('DURAT', scale='log') # PMF of age at first service, separated by outcome pmf_by_outcome('STARTAGE') # PMF of age at beginning of spell by outcome pmf_by_outcome('SPELLAGE') # PMF of age at end of spell by outcome pmf_by_outcome('EXITAGE') # PMFs of spell sequence number by outcome pmf_by_outcome('SPELL') # CDFs of time between current and next spell by outcome cdf_by_outcome('TIMER') # PMFs of gender by outcome pmf_by_outcome('GENDER') # PMFs of ethnicity by outcome pmf_by_outcome('HISPANIC') # PMFs of ethnicity by outcome pmf_by_outcome('ETHNIC')<jupyter_output><empty_output><jupyter_text>- AN - Native American - AS - Asian and Pacific - BL - Black/African American - MU - Multiple - OT - Other Category - UK - Unknown - WH - White<jupyter_code># PMFs of primary placement type by outcome pmf_by_outcome('TYPE')<jupyter_output><empty_output>
no_license
/python/spell-vis.ipynb
DataGraphito/FosterCareMobility
5
<jupyter_start><jupyter_text># A Primer on Bayesian Methods for Multilevel ModelingHierarchical or multilevel modeling is a generalization of regression modeling. *Multilevel models* are regression models in which the constituent model parameters are given **probability models**. This implies that model parameters are allowed to **vary by group**. Observational units are often naturally **clustered**. Clustering induces dependence between observations, despite random sampling of clusters and random sampling within clusters. A *hierarchical model* is a particular multilevel model where parameters are nested within one another. Some multilevel structures are not hierarchical. * e.g. "country" and "year" are not nested, but may represent separate, but overlapping, clusters of parameters We will motivate this topic using an environmental epidemiology example.### Example: Radon contamination (Gelman and Hill 2006) Let's revisit the radon contamination example from the previous section. For hierarchical modeling, we will use more of the data; we will focus on modeling radon levels in Minnesota. The EPA did the radon study in 80,000 houses. There were two important predictors: * measurement in basement or first floor (radon higher in basements) * county uranium level (positive correlation with radon levels) The hierarchy in this example is households within county. ### Data organizationFirst, we import the data from a local file, and extract Minnesota's data.<jupyter_code>%matplotlib inline import numpy as np import pandas as pd import matplotlib.pyplot as plt import seaborn as sns; sns.set_context('notebook') import warnings warnings.simplefilter("ignore") # Import radon data radon_data = pd.read_csv('../data/radon.csv', index_col=0) RANDOM_SEED = 20090425 counties = radon_data.county.unique() n_counties = counties.shape[0] county = radon_data.county_code.values log_radon = radon_data.log_radon.values floor_measure = radon_data.floor.values log_uranium = np.log(radon_data.Uppm.values) county_lookup = dict(zip(counties, np.arange(n_counties)))<jupyter_output><empty_output><jupyter_text>Distribution of radon levels in MN (log scale):<jupyter_code>radon_data.activity.apply(lambda x: np.log(x+0.1)).hist(bins=25, grid=False);<jupyter_output><empty_output><jupyter_text>## Conventional approaches The two conventional alternatives to modeling radon exposure represent the two extremes of the bias-variance tradeoff: ***Complete pooling***: Treat all counties the same, and estimate a single radon level. $$y_i = \alpha + \beta x_i + \epsilon_i$$ ***No pooling***: Model radon in each county independently. $$y_i = \alpha_{j[i]} + \beta x_i + \epsilon_i$$ where $j = 1,\ldots,85$ The errors $\epsilon_i$ may represent measurement error, temporal within-house variation, or variation among houses.Here are the point estimates of the slope and intercept for the complete pooling model:<jupyter_code>from pymc3 import Model, sample, Normal, HalfCauchy, Uniform floor = radon_data.floor.values log_radon = radon_data.log_radon.values with Model() as pooled_model: β = Normal('β', 0, sigma=10, shape=2) σ = HalfCauchy('σ', 5) θ = β[0] + β[1]*floor y = Normal('y', θ, sigma=σ, observed=log_radon) <jupyter_output><empty_output><jupyter_text>Before sampling, it is useful to conduct **prior predictive checks** to ensure that our priors are appropriate, and do not unduly constrain inference.<jupyter_code>from pymc3 import sample_prior_predictive with pooled_model: prior_checks = sample_prior_predictive(samples=1000) plt.plot( [0, 1], [prior_checks["β"][:, 0], prior_checks["β"][:, 1]], "ok", alpha=0.2) plt.xlabel("Floor measurement location") plt.xticks([0,1], ["Basement", "Floor"]) plt.ylabel("Mean log radon level");<jupyter_output><empty_output><jupyter_text>These look fine--they allow for any radon values that we would reasonably expect to see in the data.<jupyter_code>with pooled_model: pooled_trace = sample(1000, tune=1000, cores=2) b0, m0 = pooled_trace['β'].mean(axis=0) plt.scatter(radon_data.floor, np.log(radon_data.activity+0.1)) xvals = np.linspace(-0.2, 1.2) plt.plot(xvals, m0*xvals+b0, 'r--');<jupyter_output><empty_output><jupyter_text>Estimates of county radon levels for the unpooled model:<jupyter_code>with Model() as unpooled_model: β0 = Normal('β0', 0, sigma=10, shape=n_counties) β1 = Normal('β1', 0, sigma=10) σ = HalfCauchy('σ', 5) θ = β0[county] + β1*floor y = Normal('y', θ, sigma=σ, observed=log_radon) with unpooled_model: unpooled_trace = sample(1000, tune=1000, cores=2, random_seed=RANDOM_SEED) from arviz import plot_forest (ax,) = plot_forest(unpooled_trace, var_names=['β0']) ax.set_yticklabels([""]); unpooled_estimates = pd.Series(unpooled_trace['β0'].mean(axis=0), index=counties) unpooled_se = pd.Series(unpooled_trace['β0'].std(axis=0), index=counties)<jupyter_output><empty_output><jupyter_text>We can plot the ordered estimates to identify counties with high radon levels:<jupyter_code>order = unpooled_estimates.sort_values().index plt.scatter(range(len(unpooled_estimates)), unpooled_estimates[order]) for i, m, se in zip(range(len(unpooled_estimates)), unpooled_estimates[order], unpooled_se[order]): plt.plot([i,i], [m-se, m+se], 'b-') plt.xlim(-1,86); plt.ylim(-1,4) plt.ylabel('Radon estimate');plt.xlabel('Ordered county');<jupyter_output><empty_output><jupyter_text>Here are visual comparisons between the pooled and unpooled estimates for a subset of counties representing a range of sample sizes.<jupyter_code>sample_counties = ('LAC QUI PARLE', 'AITKIN', 'KOOCHICHING', 'DOUGLAS', 'CLAY', 'STEARNS', 'RAMSEY', 'ST LOUIS') fig, axes = plt.subplots(2, 4, figsize=(12, 6), sharey=True, sharex=True) axes = axes.ravel() m = unpooled_trace['β1'].mean() for i,c in enumerate(sample_counties): y = radon_data.log_radon[radon_data.county==c] x = radon_data.floor[radon_data.county==c] axes[i].scatter(x + np.random.randn(len(x))*0.01, y, alpha=0.4) # No pooling model b = unpooled_estimates[c] # Plot both models and data xvals = np.linspace(-0.2, 1.2) axes[i].plot(xvals, m*xvals+b) axes[i].plot(xvals, m0*xvals+b0, 'r--') axes[i].set_xticks([0,1]) axes[i].set_xticklabels(['basement', 'floor']) axes[i].set_ylim(-1, 3) axes[i].set_title(c) if not i%2: axes[i].set_ylabel('log radon level')<jupyter_output><empty_output><jupyter_text>Neither of these models are satisfactory: * if we are trying to identify high-radon counties, pooling is useless * we do not trust extreme unpooled estimates produced by models using few observations## Multilevel and hierarchical models When we pool our data, we imply that they are sampled from the same model. This ignores any variation among sampling units (other than sampling variance): ![pooled](images/pooled_model.png) When we analyze data unpooled, we imply that they are sampled independently from separate models. At the opposite extreme from the pooled case, this approach claims that differences between sampling units are to large to combine them: ![unpooled](images/unpooled_model.png) In a hierarchical model, parameters are viewed as a sample from a population distribution of parameters. Thus, we view them as being neither entirely different or exactly the same. This is ***parital pooling***. ![hierarchical](images/partially_pooled_model.png)We can use PyMC to easily specify multilevel models, and fit them using Markov chain Monte Carlo.## Partial pooling model The simplest partial pooling model for the household radon dataset is one which simply estimates radon levels, without any predictors at any level. A partial pooling model represents a compromise between the pooled and unpooled extremes, approximately a weighted average (based on sample size) of the unpooled county estimates and the pooled estimates. $$\hat{\alpha} \approx \frac{(n_j/\sigma_y^2)\bar{y}_j + (1/\sigma_{\alpha}^2)\bar{y}}{(n_j/\sigma_y^2) + (1/\sigma_{\alpha}^2)}$$ Estimates for counties with smaller sample sizes will shrink towards the state-wide average. Estimates for counties with larger sample sizes will be closer to the unpooled county estimates.<jupyter_code>with Model() as partial_pooling: # Priors μ_a = Normal('μ_a', mu=0., sigma=10) σ_a = HalfCauchy('σ_a', 5) # Random intercepts a = Normal('a', mu=μ_a, sigma=σ_a, shape=n_counties) # Model error σ_y = HalfCauchy('σ_y',5) # Expected value y_hat = a[county] # Data likelihood y_like = Normal('y_like', mu=y_hat, sigma=σ_y, observed=log_radon) with partial_pooling: partial_pooling_trace = sample(1000, tune=1000, cores=2, random_seed=RANDOM_SEED) sample_trace = partial_pooling_trace['a'] fig, axes = plt.subplots(1, 2, figsize=(14,6), sharex=True, sharey=True) n_samples, n_counties = sample_trace.shape jitter = np.random.normal(scale=0.1, size=n_counties) n_county = radon_data.groupby('county')['county_code'].count() unpooled_means = radon_data.groupby('county')['log_radon'].mean() unpooled_sd = radon_data.groupby('county')['log_radon'].std() unpooled = pd.DataFrame({'n':n_county, 'm':unpooled_means, 'sd':unpooled_sd}) unpooled['se'] = unpooled.sd/np.sqrt(unpooled.n) axes[0].plot(unpooled.n + jitter, unpooled.m, 'b.') for j, row in zip(jitter, unpooled.iterrows()): name, dat = row axes[0].plot([dat.n+j,dat.n+j], [dat.m-dat.se, dat.m+dat.se], 'b-') axes[0].set_xscale('log') axes[0].hlines(sample_trace.mean(), 0.9, 100, linestyles='--') n_samples, n_counties = sample_trace.shape means = sample_trace.mean(axis=0) sd = sample_trace.std(axis=0) axes[1].scatter(n_county.values + jitter, means) axes[1].set_xscale('log') axes[1].set_xlim(1,100) axes[1].set_ylim(0, 3) axes[1].hlines(sample_trace.mean(), 0.9, 100, linestyles='--') for j,n,m,s in zip(jitter, n_county.values, means, sd): axes[1].plot([n+j]*2, [m-s, m+s], 'b-');<jupyter_output><empty_output><jupyter_text>Notice the difference between the unpooled and partially-pooled estimates, particularly at smaller sample sizes. The former are both more extreme and more imprecise.## Varying intercept model This model allows intercepts to vary across county, according to a random effect. $$y_i = \alpha_{j[i]} + \beta x_{i} + \epsilon_i$$ where $$\epsilon_i \sim N(0, \sigma_y^2)$$ and the intercept random effect: $$\alpha_{j[i]} \sim N(\mu_{\alpha}, \sigma_{\alpha}^2)$$ As with the the “no-pooling” model, we set a separate intercept for each county, but rather than fitting separate least squares regression models for each county, multilevel modeling **shares strength** among counties, allowing for more reasonable inference in counties with little data.<jupyter_code>with Model() as varying_intercept: # Priors μ_a = Normal('μ_a', mu=0., sigma=10) σ_a = HalfCauchy('σ_a', 5) # Random intercepts a = Normal('a', mu=μ_a, sigma=σ_a, shape=n_counties) # Common slope b = Normal('b', mu=0., sigma=10) # Model error sd_y = HalfCauchy('sd_y', 5) # Expected value y_hat = a[county] + b * floor_measure # Data likelihood y_like = Normal('y_like', mu=y_hat, sigma=sd_y, observed=log_radon) with varying_intercept: varying_intercept_trace = sample(1000, tune=1000, cores=2, random_seed=RANDOM_SEED) (ax,) = plot_forest(varying_intercept_trace, var_names=['a']) ax.set_yticklabels([""]); from arviz import plot_posterior plot_posterior(varying_intercept_trace, var_names=['σ_a', 'b']);<jupyter_output><empty_output><jupyter_text>The estimate for the `floor` coefficient is approximately -0.66, which can be interpreted as houses without basements having about half ($\exp(-0.66) = 0.52$) the radon levels of those with basements, after accounting for county.<jupyter_code>from pymc3 import summary summary(varying_intercept_trace, var_names=['b']) xvals = np.arange(2) bp = varying_intercept_trace[a].mean(axis=0) mp = varying_intercept_trace[b].mean() for bi in bp: plt.plot(xvals, mp*xvals + bi, 'bo-', alpha=0.4) plt.xlim(-0.1,1.1);<jupyter_output><empty_output><jupyter_text>It is easy to show that the partial pooling model provides more objectively reasonable estimates than either the pooled or unpooled models, at least for counties with small sample sizes.<jupyter_code>fig, axes = plt.subplots(2, 4, figsize=(12, 6), sharey=True, sharex=True) axes = axes.ravel() for i,c in enumerate(sample_counties): # Plot county data y = radon_data.log_radon[radon_data.county==c] x = radon_data.floor[radon_data.county==c] axes[i].scatter(x + np.random.randn(len(x))*0.01, y, alpha=0.4) # No pooling model b = unpooled_estimates[c] m = unpooled_trace['β1'].mean() xvals = np.linspace(-0.2, 1.2) # Unpooled estimate axes[i].plot(xvals, m*xvals+b) # Pooled estimate axes[i].plot(xvals, m0*xvals+b0, 'r--') # Partial pooling esimate axes[i].plot(xvals, mp*xvals+bp[county_lookup[c]], 'k:') axes[i].set_xticks([0,1]) axes[i].set_xticklabels(['basement', 'floor']) axes[i].set_ylim(-1, 3) axes[i].set_title(c) if not i%2: axes[i].set_ylabel('log radon level');<jupyter_output><empty_output><jupyter_text>## Varying slope model Alternatively, we can posit a model that allows the counties to vary according to how the location of measurement (basement or floor) influences the radon reading. $$y_i = \alpha + \beta_{j[i]} x_{i} + \epsilon_i$$ <jupyter_code>with Model() as varying_slope: # Priors μ_b = Normal('μ_b', mu=0., sigma=10) σ_b = HalfCauchy('σ_b', 5) # Common intercepts a = Normal('a', mu=0., sigma=10) # Random slopes b = Normal('b', mu=μ_b, sigma=σ_b, shape=n_counties) # Model error σ_y = HalfCauchy('σ_y',5) # Expected value y_hat = a + b[county] * floor_measure # Data likelihood y_like = Normal('y_like', mu=y_hat, sigma=σ_y, observed=log_radon) with varying_slope: varying_slope_trace = sample(1000, tune=1000, cores=2, random_seed=RANDOM_SEED) (ax,) = plot_forest(varying_slope_trace, var_names=['b']) ax.set_yticklabels([""]); xvals = np.arange(2) b = varying_slope_trace['a'].mean() m = varying_slope_trace['b'].mean(axis=0) for mi in m: plt.plot(xvals, mi*xvals + b, 'bo-', alpha=0.4) plt.xlim(-0.2, 1.2);<jupyter_output><empty_output><jupyter_text>## Exercise: Varying intercept and slope model The most general model allows both the intercept and slope to vary by county: $$y_i = \alpha_{j[i]} + \beta_{j[i]} x_{i} + \epsilon_i$$ Combine these two models to create a version with both slope and intercept varying.<jupyter_code>with Model() as varying_intercept_slope: # Write your model here with varying_intercept_slope: varying_intercept_slope_trace = sample(1000, tune=1000, cores=2, random_seed=RANDOM_SEED)<jupyter_output><empty_output><jupyter_text>Forest plot of slopes and intercepts:<jupyter_code>plot_forest(varying_intercept_slope_trace, var_names=['a','b']); xvals = np.arange(2) b = varying_intercept_slope_trace['a'].mean(axis=0) m = varying_intercept_slope_trace['b'].mean(axis=0) for bi,mi in zip(b,m): plt.plot(xvals, mi*xvals + bi, 'bo-', alpha=0.4) plt.xlim(-0.1, 1.1);<jupyter_output><empty_output><jupyter_text>## Adding group-level predictors A primary strength of multilevel models is the ability to handle predictors on multiple levels simultaneously. If we consider the varying-intercepts model above: $$y_i = \alpha_{j[i]} + \beta x_{i} + \epsilon_i$$ we may, instead of a simple random effect to describe variation in the expected radon value, specify another regression model with a county-level covariate. Here, we use the county uranium reading $u_j$, which is thought to be related to radon levels: $$\alpha_j = \gamma_0 + \gamma_1 u_j + \zeta_j$$ $$\zeta_j \sim N(0, \sigma_{\alpha}^2)$$ Thus, we are now incorporating a house-level predictor (floor or basement) as well as a county-level predictor (uranium). Note that the model has both indicator variables for each county, plus a county-level covariate. In classical regression, this would result in collinearity. In a multilevel model, the partial pooling of the intercepts towards the expected value of the group-level linear model avoids this. Group-level predictors also serve to reduce group-level variation $\sigma_{\alpha}$. An important implication of this is that the group-level estimate induces stronger pooling.<jupyter_code>from pymc3 import Deterministic with Model() as hierarchical_intercept: # Priors σ_a = HalfCauchy('σ_a', 5) # County uranium model γ_0 = Normal('γ_0', mu=0., sigma=10) γ_1 = Normal('γ_1', mu=0., sigma=10) # Uranium model for intercept μ_a = γ_0 + γ_1*log_uranium # County variation not explained by uranium ϵ_a = Normal('ϵ_a', mu=0, sigma=1, shape=n_counties) a = Deterministic('a', μ_a + σ_a*ϵ_a[county]) # Common slope b = Normal('b', mu=0., sigma=10) # Model error σ_y = Uniform('σ_y', lower=0, upper=100) # Expected value y_hat = a + b * floor_measure # Data likelihood y_like = Normal('y_like', mu=y_hat, sigma=σ_y, observed=log_radon) with hierarchical_intercept: hierarchical_intercept_trace = sample(1000, tune=1000, cores=2, random_seed=RANDOM_SEED) a_means = hierarchical_intercept_trace['a'].mean(axis=0) plt.scatter(log_uranium, a_means) g0 = hierarchical_intercept_trace['γ_0'].mean() g1 = hierarchical_intercept_trace['γ_1'].mean() xvals = np.linspace(-1, 0.8) plt.plot(xvals, g0+g1*xvals, 'k--') plt.xlim(-1, 0.8) a_se = hierarchical_intercept_trace['a'].std(axis=0) for ui, m, se in zip(log_uranium, a_means, a_se): plt.plot([ui,ui], [m-se, m+se], 'b-') plt.xlabel('County-level uranium'); plt.ylabel('Intercept estimate');<jupyter_output><empty_output><jupyter_text>The standard errors on the intercepts are narrower than for the partial-pooling model without a county-level covariate.### Correlations among levels In some instances, having predictors at multiple levels can reveal correlation between individual-level variables and group residuals. We can account for this by including the average of the individual predictors as a covariate in the model for the group intercept. $$\alpha_j = \gamma_0 + \gamma_1 u_j + \gamma_2 \bar{x} + \zeta_j$$ These are broadly referred to as ***contextual effects***.<jupyter_code># Create new variable for mean of floor across counties xbar = radon_data.groupby('county')['floor'].mean().rename(county_lookup).values with Model() as contextual_effect: # Priors σ_a = HalfCauchy('σ_a', 5) # County uranium model for slope γ = Normal('γ', mu=0., sigma=10, shape=3) # Uranium model for intercept μ_a = Deterministic('μ_a', γ[0] + γ[1]*log_uranium + γ[2]*xbar[county]) # County variation not explained by uranium ϵ_a = Normal('ϵ_a', mu=0, sigma=1, shape=n_counties) a = Deterministic('a', μ_a + σ_a*ϵ_a[county]) # Common slope b = Normal('b', mu=0., sigma=10) # Model error σ_y = Uniform('σ_y', lower=0, upper=100) # Expected value y_hat = a + b * floor_measure # Data likelihood y_like = Normal('y_like', mu=y_hat, sigma=σ_y, observed=log_radon) with contextual_effect: contextual_effect_trace = sample(1000, tune=1000, cores=2, random_seed=RANDOM_SEED) plot_forest(contextual_effect_trace, var_names=['γ']); summary(contextual_effect_trace, var_names=['γ'])<jupyter_output><empty_output><jupyter_text>So, we might infer from this that counties with higher proportions of houses without basements tend to have higher baseline levels of radon. Perhaps this is related to the soil type, which in turn might influence what type of structures are built.### Prediction Gelman (2006) used cross-validation tests to check the prediction error of the unpooled, pooled, and partially-pooled models **root mean squared cross-validation prediction errors**: * unpooled = 0.86 * pooled = 0.84 * multilevel = 0.79 There are two types of prediction that can be made in a multilevel model: 1. a new individual within an existing group 2. a new individual within a new group For example, if we wanted to make a prediction for a new house with no basement in St. Louis county, we just need to sample from the radon model with the appropriate intercept.<jupyter_code>county_lookup['ST LOUIS']<jupyter_output><empty_output><jupyter_text>That is, $$\tilde{y}_i \sim N(\alpha_{69} + \beta (x_i=1), \sigma_y^2)$$ This is a matter of adding a single additional line in PyMC:<jupyter_code>with Model() as contextual_pred: # Priors σ_a = HalfCauchy('σ_a', 5) # County uranium model for slope γ = Normal('γ', mu=0., sigma=10, shape=3) # Uranium model for intercept μ_a = Deterministic('μ_a', γ[0] + γ[1]*log_uranium + γ[2]*xbar[county]) # County variation not explained by uranium ϵ_a = Normal('ϵ_a', mu=0, sigma=1, shape=n_counties) a = Deterministic('a', μ_a + σ_a*ϵ_a[county]) # Common slope b = Normal('b', mu=0., sigma=10) # Model error σ_y = Uniform('σ_y', lower=0, upper=100) # Expected value y_hat = a + b * floor_measure # Data likelihood y_like = Normal('y_like', mu=y_hat, sigma=σ_y, observed=log_radon) # St Louis county prediction stl_pred = Normal('stl_pred', mu=a[69] + b, sigma=σ_y) with contextual_pred: contextual_pred_trace = sample(2000, tune=1000, cores=2, random_seed=RANDOM_SEED) plot_posterior(contextual_pred_trace, var_names=['stl_pred']);<jupyter_output><empty_output><jupyter_text>## Exercise How would we make a prediction from a new county (*e.g.* one not included in this dataset)?<jupyter_code># Write your answer here<jupyter_output><empty_output>
permissive
/notebooks/Section1_2-Hierarchical_Models.ipynb
Robert-Muil/bayes_course_july2020
20
<jupyter_start><jupyter_text># Requesting Argo BGC data from Ifremer erddap, expert mode Using the expert mode, you can have access to all fields retrieved from the erddap, including all QC variables and without any data mode filtering. *** Script prepared by [Guillaume Maze](http://github.com/gmaze) (Mar. 2020)<jupyter_code>import sys, os import numpy as np import xarray as xr try: import argopy except ModuleNotFoundError: !pip install git+http://github.com/euroargodev/argopy.git@master import argopy print("argopy:", argopy.__version__) from argopy import DataFetcher as ArgoDataFetcher # import matplotlib.pyplot as plt import matplotlib as mpl import matplotlib.cm as cm import cmocean from pandas.plotting import register_matplotlib_converters register_matplotlib_converters() %matplotlib inline # Usefull colormaps and colorbar makers: qcmap = mpl.colors.ListedColormap(['#000000', '#31FC03', '#ADFC03', '#FCBA03', '#FC1C03', '#324CA8', '#000000', '#000000', '#B22CC9', '#000000']) def colorbar_qc(cmap, **kwargs): """Adjust colorbar ticks with discrete colors for QC flags""" ncolors = 10 mappable = cm.ScalarMappable(cmap=cmap) mappable.set_array([]) mappable.set_clim(-0.5, ncolors+0.5) colorbar = plt.colorbar(mappable, **kwargs) colorbar.set_ticks(np.linspace(0, ncolors, ncolors)) colorbar.set_ticklabels(range(ncolors)) return colorbar from sklearn import preprocessing le = preprocessing.LabelEncoder() le.fit(['A','D','R']) dmode_map = mpl.colors.ListedColormap(['#FCBA03','#31FC03','#FF0000']) def colorbar_dmode(cmap=dmode_map, **kwargs): """Adjust colorbar ticks with discrete colors for DATA MODE""" ncolors = 3 mappable = cm.ScalarMappable(cmap=cmap) mappable.set_array([]) mappable.set_clim(-0.5, ncolors+0.5) colorbar = plt.colorbar(mappable, **kwargs) colorbar.set_ticks(np.linspace(0, ncolors, ncolors)) colorbar.set_ticklabels(le.classes_) return colorbar<jupyter_output>argopy: 0.1.1 <jupyter_text># Create the Argo data loader instance If we want to retrieve data without post-processing, we need to specify the ``mode`` option to ``expert`` when creating the data loader instance.<jupyter_code># argo_loader = ArgoDataFetcher(mode='expert') # argo_loader = ArgoDataFetcher(mode='expert', cachedir='tmp') argo_loader = ArgoDataFetcher(mode='expert', ds='bgc') argo_loader<jupyter_output><empty_output><jupyter_text># Example of data fetching for a specific region (Upper Equatorial East-Pacific) **NOTE**: if the following cell is raising a "Internal Server Error", simply try to reload it (this is due to a long (>1min) delay in preparing the response)<jupyter_code># box = [-120., -85., -10, 10, 0, 1000, '2018-01-01','2018-12-31'] # box = [-120., -85., -10, 10, 0, 500, '2019-01-01','2019-12-31'] box = [-120., -85., -10, 10, 0, 500, '2017-01-01','2017-12-31'] # box = [-120., -85., -30, -10, 0, 500, '2019-01-01','2019-12-31'] ds = argo_loader.region(box).to_xarray() ds<jupyter_output><empty_output><jupyter_text>## QC figure with unfiltered data<jupyter_code>fig, ax = plt.subplots(nrows=3, ncols=2, figsize=(25,20)) ax = np.array(ax).flatten() fig.delaxes(ax[-1]) ix = 0 sc = ax[ix].scatter(ds['LONGITUDE'], ds['LATITUDE'], c=ds['POSITION_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].set_xlabel('LONGITUDE') ax[ix].set_ylabel('LATITUDE') ax[ix].set_title('POSITION_QC') ix += 1 sc = ax[ix].scatter(ds['TIME'].values, ds['CYCLE_NUMBER'], c=ds['TIME_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].set_xlabel('TIME') ax[ix].set_ylabel('Cycle number') ax[ix].set_title('TIME_QC') ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['TEMP_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel('CYCLE_NUMBER') ax[ix].set_ylabel('Pressure') ax[ix].set_title('TEMP_QC') ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['PSAL_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel('CYCLE_NUMBER') ax[ix].set_ylabel('Pressure') ax[ix].set_title('PSAL_QC') ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['DOXY_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel('CYCLE_NUMBER') ax[ix].set_ylabel('Pressure') ax[ix].set_title('DOXY_QC') fig.suptitle("RAW variables QC\n%s" % ds.attrs['Fetched_constraints']);<jupyter_output><empty_output><jupyter_text># Filter data according to data mode But this is (nearly) the raw output of the request. It could be useful to simply select the most appropriate variables according to the data mode (Real time, adjusted and delayed mode). Note that this is done automatically in ``mode='standard'``.<jupyter_code># Inspect data mode fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(20,5)) ax = np.array(ax).flatten() sc = ax[0].scatter(ds['TIME'].values, ds['CYCLE_NUMBER'], c=le.transform(ds['DATA_MODE']), vmin=0, vmax=2, cmap=dmode_map) colorbar_dmode(label='DATA_MODE', ax=ax[0]) ax[0].set_xlabel('TIME') ax[0].set_ylabel('Cycle number') ax[0].grid() ax[0].set_title(ds['DATA_MODE'].attrs['convention']);<jupyter_output><empty_output><jupyter_text>## Filter data according to data mode<jupyter_code>ds_filtered = ds.argo.filter_data_mode() ds_filtered<jupyter_output><empty_output><jupyter_text>## QC figure with appropriate variables<jupyter_code>fig, ax = plt.subplots(nrows=2, ncols=2, figsize=(20,10)) ax = np.array(ax).flatten() ix = 0 sc = ax[ix].scatter(ds_filtered['LONGITUDE'], ds_filtered['LATITUDE'], c=ds_filtered['POSITION_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].set_xlabel('LONGITUDE') ax[ix].set_ylabel('LATITUDE') ax[ix].set_title('POSITION_QC') ix += 1 sc = ax[ix].scatter(ds_filtered['TIME'].values, ds_filtered['CYCLE_NUMBER'], c=ds_filtered['TIME_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].set_xlabel('TIME') ax[ix].set_ylabel('Cycle number') ax[ix].set_title('TIME_QC') ix += 1 sc = ax[ix].scatter(ds_filtered['CYCLE_NUMBER'], ds_filtered['PRES'], c=ds_filtered['TEMP_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel('CYCLE_NUMBER') ax[ix].set_ylabel('Pressure') ax[ix].set_title('TEMP_QC') ix += 1 sc = ax[ix].scatter(ds_filtered['CYCLE_NUMBER'], ds_filtered['PRES'], c=ds_filtered['PSAL_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel('CYCLE_NUMBER') ax[ix].set_ylabel('Pressure') ax[ix].set_title('PSAL_QC') fig.suptitle("QC flags of Data mode filtered variables\n%s" % ds.attrs['Fetched_constraints']);<jupyter_output><empty_output><jupyter_text># Example of data fetching for a specific float<jupyter_code>float_fetcher = argo_loader.float(3901530) # float_fetcher = argo_loader.float(6902746) # float_fetcher = argo_loader.region([-180,180,0,90,0,1000]) # float_fetcher = argo_loader.float(2903005) # Float with a RBRargo sensor float_fetcher ds_unfiltered = float_fetcher.to_xarray() ds = ds_unfiltered.argo.filter_data_mode() ds # Get basic information about the float: print('This float has performed: %i profiles' % len(np.unique(ds['CYCLE_NUMBER']))) print('This float operated between %s and %s' % (ds['TIME'].min().values, ds['TIME'].max().values)) print('This float profiles range from %0.1fdb to %0.1fdb' % (ds['PRES'].min(), ds['PRES'].max() ))<jupyter_output><empty_output><jupyter_text>## Plot trajectory<jupyter_code>this = ds.reset_coords().groupby('CYCLE_NUMBER').min() plt.plot(this['LONGITUDE'], this['LATITUDE'], '-', color=[0.7]*3, zorder=0) plt.scatter(this['LONGITUDE'], this['LATITUDE'], c=this['CYCLE_NUMBER']) plt.xlabel('LONGITUDE') plt.ylabel('LATITUDE') plt.gca().grid() plt.colorbar(label='CYCLE_NUMBER') plt.title(ds.attrs['Fetched_constraints']) plt.show()<jupyter_output><empty_output><jupyter_text>## QC flags for appropriate variables<jupyter_code>fig, ax = plt.subplots(nrows=2, ncols=2, figsize=(20,10)) ax = np.array(ax).flatten() ix = 0 ax[ix].plot(ds['LONGITUDE'], ds['LATITUDE'], '-', color=[0.7]*3, zorder=0) sc = ax[ix].scatter(ds['LONGITUDE'], ds['LATITUDE'], c=ds['POSITION_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].set_xlabel('LONGITUDE') ax[ix].set_ylabel('LATITUDE') ax[ix].set_title('POSITION_QC') ix += 1 sc = ax[ix].scatter(ds['TIME'].values, ds['CYCLE_NUMBER'], c=ds['TIME_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].set_xlabel('TIME') ax[ix].set_ylabel('Cycle number') ax[ix].set_title('TIME_QC') ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['TEMP_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel('CYCLE_NUMBER') ax[ix].set_ylabel('Pressure') ax[ix].set_title('TEMP_QC') ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['PSAL_QC'], vmin=0, vmax=9, cmap=qcmap) colorbar_qc(qcmap, ax=ax[ix]) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel('CYCLE_NUMBER') ax[ix].set_ylabel('Pressure') ax[ix].set_title('PSAL_QC') fig.suptitle("Data mode filtered variables QC\n%s" % ds.attrs['Fetched_constraints']);<jupyter_output><empty_output><jupyter_text>## DATA MODE<jupyter_code>fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(10,5)) ax = np.array(ax).flatten() sc = ax[0].scatter(ds['TIME'].values, ds['CYCLE_NUMBER'], c=le.transform(ds['DATA_MODE']), vmin=0, vmax=2, cmap=dmode_map) colorbar_dmode(label='DATA_MODE', ax=ax[0]) ax[0].set_ylabel('CYCLE_NUMBER') ax[0].set_xlabel('TIME') ax[0].grid() ax[0].set_title(ds['DATA_MODE'].attrs['convention']);<jupyter_output><empty_output><jupyter_text>## Hovmoller<jupyter_code>fig, ax = plt.subplots(nrows=3, ncols=1, figsize=(20,10), sharex=True, sharey=True) ax = np.array(ax).flatten() ix = 0 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['TEMP'], cmap=cmocean.cm.thermal) plt.colorbar(sc, ax=ax[ix], label=ds['TEMP'].attrs['units']) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['TEMP'].attrs['long_name']) ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['PSAL'], cmap=cmocean.cm.haline) plt.colorbar(sc, ax=ax[ix], label=ds['PSAL'].attrs['units']) ax[ix].grid() ax[ix].set_xlabel(ds['CYCLE_NUMBER'].attrs['long_name']) ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['PSAL'].attrs['long_name']); ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['DOXY'], cmap=cmocean.cm.algae) plt.colorbar(sc, ax=ax[ix], label=ds['DOXY'].attrs['units']) ax[ix].grid() ax[ix].set_xlabel(ds['CYCLE_NUMBER'].attrs['long_name']) ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['DOXY'].attrs['long_name']); fig.suptitle("%s\n(all data, no QC filter)" % ds.attrs['Fetched_constraints'], fontsize=14); fig, ax = plt.subplots(nrows=2, ncols=2, figsize=(20,10), sharex=True, sharey=True) ax = np.array(ax).flatten() ix = 0 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['TEMP'], cmap=cmocean.cm.thermal) plt.colorbar(sc, ax=ax[ix], label=ds['TEMP'].attrs['units']) ax[ix].invert_yaxis() ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['TEMP'].attrs['long_name']) ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['TEMP_ERROR'], cmap=cmocean.cm.amp) plt.colorbar(sc, ax=ax[ix], label=ds['TEMP_ERROR'].attrs['units']) ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['TEMP_ERROR'].attrs['long_name']) ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['PSAL'], cmap=cmocean.cm.haline) plt.colorbar(sc, ax=ax[ix], label=ds['PSAL'].attrs['units']) ax[ix].set_xlabel(ds['CYCLE_NUMBER'].attrs['long_name']) ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['PSAL'].attrs['long_name']); ix += 1 sc = ax[ix].scatter(ds['CYCLE_NUMBER'], ds['PRES'], c=ds['PSAL_ERROR'], cmap=cmocean.cm.amp) plt.colorbar(sc, ax=ax[ix], label=ds['PSAL_ERROR'].attrs['units']) ax[ix].set_xlabel(ds['CYCLE_NUMBER'].attrs['long_name']) ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['PSAL_ERROR'].attrs['long_name']);<jupyter_output><empty_output><jupyter_text>## Super-imposed profiles<jupyter_code>fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(20,10), sharey=True) ax = np.array(ax).flatten() ix = 0 sc = ax[ix].scatter(ds['TEMP'], ds['PRES'], c=ds['CYCLE_NUMBER']) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel(ds['TEMP'].attrs['units']) ax[ix].set_ylabel("%s [%s]" % (ds['PRES'].attrs['long_name'], ds['PRES'].attrs['units'])) ax[ix].set_title(ds['TEMP'].attrs['long_name']) ix += 1 sc = ax[ix].scatter(ds['PSAL'], ds['PRES'], c=ds['CYCLE_NUMBER']) plt.colorbar(sc, ax=ax[ix], label=ds['CYCLE_NUMBER'].attrs['long_name']) ax[ix].grid() ax[ix].set_xlabel(ds['PSAL'].attrs['units']) ax[ix].set_title(ds['PSAL'].attrs['long_name']); <jupyter_output><empty_output><jupyter_text>## T/S diagram<jupyter_code>fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(6,6)) ax = np.array(ax).flatten() ix = 0 sc = ax[ix].scatter(ds['PSAL'], ds['TEMP'], c=ds['CYCLE_NUMBER']) ax[ix].grid() ax[ix].set_xlabel("%s [%s]" % (ds['PSAL'].attrs['long_name'], ds['PSAL'].attrs['units'])) ax[ix].set_ylabel("%s [%s]" % (ds['TEMP'].attrs['long_name'], ds['TEMP'].attrs['units'])) plt.title(ds.attrs['Fetched_constraints']);<jupyter_output><empty_output><jupyter_text># Filter data according to QC flags<jupyter_code># ds_ok = float_fetcher.fetcher.filter_qc(ds) # By default, drop all points with QC not equal to 1 or 2 in all variables ds_ok = float_fetcher.fetcher.filter_qc(ds, mode='any') # Keep points with at least of the variable QC in 1 or 2 ds_ok fig, ax = plt.subplots(nrows=2, ncols=2, figsize=(20,10), sharex=True, sharey=True) ax = np.array(ax).flatten() ix = 0 sc = ax[ix].scatter(ds_ok['CYCLE_NUMBER'], ds_ok['PRES'], c=ds_ok['TEMP'], cmap=cmocean.cm.thermal) plt.colorbar(sc, ax=ax[ix], label=ds_ok['TEMP'].attrs['units']) ax[ix].invert_yaxis() ax[ix].set_ylabel("%s [%s]" % (ds_ok['PRES'].attrs['long_name'], ds_ok['PRES'].attrs['units'])) ax[ix].set_title(ds_ok['TEMP'].attrs['long_name']) ix += 1 sc = ax[ix].scatter(ds_ok['CYCLE_NUMBER'], ds_ok['PRES'], c=ds_ok['TEMP_ERROR'], cmap=cmocean.cm.amp) plt.colorbar(sc, ax=ax[ix], label=ds_ok['TEMP_ERROR'].attrs['units']) ax[ix].set_ylabel("%s [%s]" % (ds_ok['PRES'].attrs['long_name'], ds_ok['PRES'].attrs['units'])) ax[ix].set_title(ds_ok['TEMP_ERROR'].attrs['long_name']) ix += 1 sc = ax[ix].scatter(ds_ok['CYCLE_NUMBER'], ds_ok['PRES'], c=ds_ok['PSAL'], cmap=cmocean.cm.haline) plt.colorbar(sc, ax=ax[ix], label=ds_ok['PSAL'].attrs['units']) ax[ix].set_xlabel(ds_ok['CYCLE_NUMBER'].attrs['long_name']) ax[ix].set_ylabel("%s [%s]" % (ds_ok['PRES'].attrs['long_name'], ds_ok['PRES'].attrs['units'])) ax[ix].set_title(ds_ok['PSAL'].attrs['long_name']); ix += 1 sc = ax[ix].scatter(ds_ok['CYCLE_NUMBER'], ds_ok['PRES'], c=ds_ok['PSAL_ERROR'], cmap=cmocean.cm.amp) plt.colorbar(sc, ax=ax[ix], label=ds_ok['PSAL_ERROR'].attrs['units']) ax[ix].set_xlabel(ds_ok['CYCLE_NUMBER'].attrs['long_name']) ax[ix].set_ylabel("%s [%s]" % (ds_ok['PRES'].attrs['long_name'], ds_ok['PRES'].attrs['units'])) ax[ix].set_title(ds_ok['PSAL_ERROR'].attrs['long_name']); fig.suptitle("Data mode and QC filtered variables\n%s" % ds.attrs['Fetched_constraints']); fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(20,10), sharey=True) ax = np.array(ax).flatten() ix = 0 sc = ax[ix].scatter(ds_ok['TEMP'], ds_ok['PRES'], c=ds_ok['CYCLE_NUMBER']) ax[ix].grid() ax[ix].invert_yaxis() ax[ix].set_xlabel(ds_ok['TEMP'].attrs['units']) ax[ix].set_ylabel("%s [%s]" % (ds_ok['PRES'].attrs['long_name'], ds_ok['PRES'].attrs['units'])) ax[ix].set_title(ds_ok['TEMP'].attrs['long_name']) ix += 1 sc = ax[ix].scatter(ds_ok['PSAL'], ds_ok['PRES'], c=ds_ok['CYCLE_NUMBER']) plt.colorbar(sc, ax=ax[ix], label=ds_ok['CYCLE_NUMBER'].attrs['long_name']) ax[ix].grid() ax[ix].set_xlabel(ds_ok['PSAL'].attrs['units']) ax[ix].set_title(ds_ok['PSAL'].attrs['long_name']); fig.suptitle("Data mode and QC filtered variables\n%s" % ds.attrs['Fetched_constraints']);<jupyter_output><empty_output>
non_permissive
/examples/Example-Expert-02.ipynb
euroargodev/erddap_usecases
15
<jupyter_start><jupyter_text># Gapminder上的农用地数据清理 按照练习4最后一个小练习的要求,从[Gapminder](https://www.gapminder.org/data/)上下载感兴趣的数据,并用[lesson5 探索两个变量](https://classroom.udacity.com/nanodegrees/nd002-cn-advanced-vip/parts/7af8e761-54e5-4c80-a284-e402c30a791b)上学到的基本方法绘制2-5张图。 此部分是数据的获取、清理部分。分析部分在此文件夹中的R部分完成。 整个清理步骤为:收集数据—评估数据—清理数据。 值得一提的是,由于是小练习,不准备花费过多时间,所以并不准备全面评估并解决所有问题,当数据清理达到使用目的即可。 ## 分析目的 简单浏览了数据后,确定主要来分析农用地面积比例与年份的关系。 准备制作3张图: (1)散点图。 (2) ## 收集数据 从网站上下载全球农用地数据。 **数据名称:**Agricultural land (% of land area); **数据格式:**Excel; **数据原始来源:**world bank; **数据描述:**1960年代到2009年各国的农用地面积比重。<jupyter_code>import pandas as pd df = pd.read_csv()<jupyter_output><empty_output>
no_license
/R basic/lesson4/.ipynb_checkpoints/agriculture_land_data_clean-checkpoint.ipynb
Tanghuaizhi/learn_data_analysis
1
<jupyter_start><jupyter_text># Unsupervised Anomaly Detection Brain-MRI Jupyter notebook for running all the experiments from our [paper](https://arxiv.org/abs/2004.03271). Hyperparameters may have to be adjusted!## Preperation ### Imports and installation of the required libraries <jupyter_code># from google.colab import drive # from google.colab import files import os, glob ! pip install pynrrd ! pip install SimpleITK ! pip install bunch ! pip install nibabel ! pip install medpy ! pip install opencv-python<jupyter_output><empty_output><jupyter_text>### Get Code Clone Code from github.com<jupyter_code># ! git clone https://github.com/StefanDenn3r/Unsupervised_Anomaly_Detection_Brain_MRI # ! cd Unsupervised_Anomaly_Detection_Brain_MRI/<jupyter_output><empty_output><jupyter_text>### Google Drive mount Mounting Google Drive to access datasets.<jupyter_code># drive.mount('gdrive')<jupyter_output><empty_output><jupyter_text>Check Directory<jupyter_code>os.getcwd()<jupyter_output><empty_output><jupyter_text>### Tensorboard and tunneling Install ngrok for tunneling <jupyter_code>if os.path.exists("ngrok-stable-linux-amd64.zip"): os.remove("ngrok-stable-linux-amd64.zip") if os.path.exists("ngrok"): os.remove("ngrok") !wget https://bin.equinox.io/c/4VmDzA7iaHb/ngrok-stable-linux-amd64.zip !unzip ngrok-stable-linux-amd64.zip<jupyter_output>'wget' is not recognized as an internal or external command, operable program or batch file. 'unzip' is not recognized as an internal or external command, operable program or batch file. <jupyter_text>Start tensorboard and forward port with ngrok<jupyter_code>LOG_DIR = 'logs/' get_ipython().system_raw( 'tensorboard --logdir {} --host 0.0.0.0 --port 6006 &' .format(LOG_DIR) ) get_ipython().system_raw('./ngrok http 6006 &')<jupyter_output><empty_output><jupyter_text>Extract ngrok url for accessing tensorboard **Attention**: Sometimes it throws an error like this: ``` json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0) ``` If this is the case the easiest way to solve this issue is to delete the ngrok*.zip and ngrok from the Google Drive folder and install them again. <jupyter_code>! curl -s http://localhost:4040/api/tunnels | python3 -c "import sys, json; print(json.load(sys.stdin)['tunnels'][0]['public_url'])" !pip3 install tensorflow==1.15<jupyter_output><empty_output><jupyter_text>## Training ### Imports<jupyter_code># %tensorflow_version 1.x import tensorflow.compat.v1 as tf tf.disable_v2_behavior() import json import os import tensorflow as tf import cv2 from datetime import datetime from utils.default_config_setup import get_config, get_options, get_datasets from trainers.AE import AE from trainers.VAE import VAE from trainers.CE import CE from trainers.ceVAE import ceVAE from trainers.VAE_You import VAE_You from trainers.GMVAE import GMVAE from trainers.GMVAE_spatial import GMVAE_spatial from trainers.fAnoGAN import fAnoGAN from trainers.ConstrainedAAE import ConstrainedAAE from trainers.ConstrainedAE import ConstrainedAE from trainers.AnoVAEGAN import AnoVAEGAN from models import autoencoder, variational_autoencoder, context_encoder_variational_autoencoder, variational_autoencoder_Zimmerer, context_encoder_variational_autoencoder, context_encoder_variational_autoencoder_Zimmerer, gaussian_mixture_variational_autoencoder_You, gaussian_mixture_variational_autoencoder_spatial, gaussian_mixture_variational_autoencoder, fanogan, fanogan_schlegl, constrained_autoencoder, constrained_adversarial_autoencoder, constrained_adversarial_autoencoder_Chen, anovaegan from utils import Evaluation from utils.default_config_setup import get_config, get_options, get_datasets, Dataset <jupyter_output><empty_output><jupyter_text>Set paths to datasets and where to save checkpoints and evaluations.<jupyter_code>def get_CONFIG(timestamp=None): current_time = datetime.now().strftime('%Y%m%d_%H%M%S') if timestamp: current_time=timestamp dataset_root = "C:/Users/Irfixq/Desktop/P2/Code 1/Unsupervised_Anomaly_Detection_Brain_MRI-master" save_dir = "C:/Users/Irfixq/Desktop/P2/Code 1/Unsupervised_Anomaly_Detection_Brain_MRI-master" CONFIG = { "BRAINWEBDIR": os.path.join(dataset_root, 'Brainweb'), "MSSEG2008DIR": os.path.join(dataset_root, 'MSSEG2008'), "MSISBI2015DIR": os.path.join(dataset_root, 'ISBIMSlesionChallenge'), "MSLUBDIR": os.path.join(dataset_root, 'MSlub'), "CHECKPOINTDIR": os.path.join(save_dir, 'checkpoints', current_time), "SAMPLEDIR": os.path.join(save_dir, 'sample_dir', current_time), } return CONFIG<jupyter_output><empty_output><jupyter_text>### Manual Training #### Baseline**AE**<jupyter_code>import tensorflow.compat.v1 as tf tf.disable_v2_behavior() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=AE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) # Create an instance of the model and train it model = AE(tf.Session(), config, network=autoencoder.autoencoder) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>**VAE**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=VAE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) # Create an instance of the model and train it model = VAE(tf.Session(), config, network=variational_autoencoder.variational_autoencoder) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>#### ceVAE - Variations Paper: [Context-encoding Variational Autoencoder for Unsupervised Anomaly Detection](https://arxiv.org/abs/1812.05941) **CE**<jupyter_code>tf.reset_default_graph() dataset = Dataset.Brainweb options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=CE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) # Create an instance of the model and train it model = CE(tf.Session(), config, network=autoencoder.autoencoder) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>**ceVAE**<jupyter_code>tf.reset_default_graph() dataset = Dataset.Brainweb options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=ceVAE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.use_gradient_based_restoration = 0.002 # Create an instance of the model and train it model = ceVAE(tf.Session(), config, network=context_encoder_variational_autoencoder.context_encoder_variational_autoencoder) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>**VAE-Zimmerer**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=64, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=VAE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) # Create an instance of the model and train it model = VAE(tf.Session(), config, network=variational_autoencoder_Zimmerer.variational_autoencoder_Zimmerer) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>**ceVAE-Zimmerer**<jupyter_code>tf.reset_default_graph() dataset = Dataset.Brainweb options = get_options(batchsize=64, learningrate=0.0001, numEpochs=1, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=ceVAE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) # Create an instance of the model and train it model = ceVAE(tf.Session(), config, network=context_encoder_variational_autoencoder_Zimmerer.context_encoder_variational_autoencoder_Zimmerer) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>#### GMVAE-(Restoration)-Variations Paper: [Unsupervised Lesion Detection via Image Restoration with a Normative Prior](https://openreview.net/forum?id=S1xg4W-leV) **VAE-You**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=VAE_You, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.restore_lr = 1e-3 config.restore_steps = 10 config.tv_lambda = 0.0 # Create an instance of the model and train it model = VAE_You(tf.Session(), config, network=variational_autoencoder.variational_autoencoder) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>**GMVAE-You**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=GMVAE_spatial, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.dim_c = 9 config.dim_z = 1 config.dim_w = 1 config.c_lambda = 1 config.restore_lr = 1e-3 config.restore_steps = 10 config.tv_lambda = 1 # Create an instance of the model and train it model = GMVAE_spatial(tf.Session(), config, network=gaussian_mixture_variational_autoencoder_You.gaussian_mixture_variational_autoencoder_You) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>**GMVAE**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=GMVAE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.dim_c = 9 config.dim_z = 128 config.dim_w = 1 config.c_lambda = 1 config.restore_lr = 1e-3 config.restore_steps = 10 config.tv_lambda = 0.0 # Create an instance of the model and train it model = GMVAE(tf.Session(), config, network=gaussian_mixture_variational_autoencoder.gaussian_mixture_variational_autoencoder) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs']))<jupyter_output><empty_output><jupyter_text>**GMVAE-spatial**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=GMVAE_spatial, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.dim_c = 9 config.dim_z = 1 config.dim_w = 1 config.c_lambda = 1 config.restore_lr = 1e-3 config.restore_steps = 10 config.tv_lambda = 0.0 # Create an instance of the model and train it model = GMVAE_spatial(tf.Session(), config, network=gaussian_mixture_variational_autoencoder_spatial.gaussian_mixture_variational_autoencoder_spatial) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs']))<jupyter_output><empty_output><jupyter_text>#### f-AnoGAN Paper: [f-AnoGAN: Fast unsupervised anomaly detection with generative adversarial networks.](https://www.ncbi.nlm.nih.gov/pubmed/30831356)**Unified f-AnoGan**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=fAnoGAN, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.kappa = 1.0 config.scale = 10.0 # Create an instance of the model and train it model = fAnoGAN(tf.Session(), config, network=fanogan.fanogan) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs']))<jupyter_output><empty_output><jupyter_text>**f-AnoGAN - Schlegl**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=8, learningrate=0.0001, numEpochs=2, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=fAnoGAN, options=options, optimizer='ADAM', intermediateResolutions=[16, 16], dropout_rate=0.1, dataset=datasetHC) config.kappa = 1.0 config.scale = 10.0 # Create an instance of the model and train it model = fAnoGAN(tf.Session(), config, network=fanogan_schlegl.fanogan_schlegl) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>#### Constrained Adversarial AE Paper: [Unsupervised Detection of Lesions in Brain MRI using constrained adversarial auto-encoders](https://arxiv.org/abs/1806.04972)**constrained AAE**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=ConstrainedAAE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.scale = 10.0 config.rho = 1.0 # Create an instance of the model and train it model = ConstrainedAAE(tf.Session(), config, network=constrained_adversarial_autoencoder.constrained_adversarial_autoencoder) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output><jupyter_text>**constrained AAE Chen**<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=8, learningrate=0.0001, numEpochs=2, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=ConstrainedAAE, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) config.kappa = 1.0 config.scale = 10.0 config.rho = 1.0 # Create an instance of the model and train it model = ConstrainedAAE(tf.Session(), config, network=constrained_adversarial_autoencoder_Chen.constrained_adversarial_autoencoder_Chen) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs']))<jupyter_output><empty_output><jupyter_text>#### AnoVAEGAN Paper: [Deep autoencoding models for unsupervised anomaly segmentation in brain MR images](https://arxiv.org/abs/1804.04488)<jupyter_code>tf.reset_default_graph() dataset = Dataset.BRAINWEB options = get_options(batchsize=128, learningrate=0.0001, numEpochs=20, zDim=128, outputWidth=128, outputHeight=128, config=get_CONFIG()) options['data']['dir'] = options["globals"][dataset.value] datasetHC, datasetPC = get_datasets(options, dataset) config = get_config(trainer=AnoVAEGAN, options=options, optimizer='ADAM', intermediateResolutions=[8, 8], dropout_rate=0.1, dataset=datasetHC) # Create an instance of the model and train it model = AnoVAEGAN(tf.Session(), config, network=anovaegan.anovaegan) # Train it model.train(datasetHC) # Evaluate Evaluation.evaluate(datasetPC, model, options, description=f"{type(datasetHC).__name__}-{options['threshold']}", epoch=str(options['train']['numEpochs'])) <jupyter_output><empty_output>
non_permissive
/Unsupervised Anomaly Detection Brain-MRI.ipynb
irfixq/AE
24
<jupyter_start><jupyter_text>Installing (updating) the following libraries for your Sagemaker instance.<jupyter_code>!pip install .. # installing d2l <jupyter_output><empty_output><jupyter_text># 自我注意力和位置编码 :label:`sec_self-attention-and-positional-encoding` 在深度学习中,我们经常使用 CNN 或 RNN 对序列进行编码。现在请注意机制。想象一下,我们将一系列令牌输入注意力池,以便同一组令牌充当查询、键和值。具体来说,每个查询都会关注所有键值对并生成一个注意力输出。由于查询、键和值来自同一个地方,因此执行 *自我关注 * :cite:`Lin.Feng.Santos.ea.2017,Vaswani.Shazeer.Parmar.ea.2017`,也称为 * 内心注意 * :cite:`Cheng.Dong.Lapata.2016,Parikh.Tackstrom.Das.ea.2016,Paulus.Xiong.Socher.2017`。 在本节中,我们将讨论使用自我注意的序列编码,包括使用序列顺序的其他信息。 <jupyter_code>import math import torch from torch import nn from d2l import torch as d2l<jupyter_output><empty_output><jupyter_text>## 自我注意 给定一系列输入令牌 $\mathbf{x}_1, \ldots, \mathbf{x}_n$,其中任何 $\mathbf{x}_i \in \mathbb{R}^d$ ($1 \leq i \leq n$),它的自我注意力输出一个长度相同的序列 $\mathbf{y}_1, \ldots, \mathbf{y}_n$,其中 $$\mathbf{y}_i = f(\mathbf{x}_i, (\mathbf{x}_1, \mathbf{x}_1), \ldots, (\mathbf{x}_n, \mathbf{x}_n)) \in \mathbb{R}^d$$ 根据 :eqref:`eq_attn-pooling` 中关注集中 $f$ 的定义。使用多头注意力,以下代码片段计算具有形状的张量的自我注意力(批量大小、时间步长或令牌中的序列长度,$d$)。输出张量的形状相同。 <jupyter_code>num_hiddens, num_heads = 100, 5 attention = d2l.MultiHeadAttention(num_hiddens, num_hiddens, num_hiddens, num_hiddens, num_heads, 0.5) attention.eval() batch_size, num_queries, valid_lens = 2, 4, torch.tensor([3, 2]) X = torch.ones((batch_size, num_queries, num_hiddens)) attention(X, X, X, valid_lens).shape<jupyter_output><empty_output><jupyter_text>## 比较 CNN、RNN 和自我注意 :label:`subsec_cnn-rnn-self-attention` 让我们比较将 $n$ 令牌序列映射到另一个相等长度序列的架构,其中每个输入或输出令牌由 $d$ 维矢量表示。具体来说,我们将考虑 CNN、RNN 和自我注意力。我们将比较它们的计算复杂性、顺序操作和最大路径长度。请注意,顺序操作会阻止并行计算,而任意序列位置组合之间的路径较短,可以更轻松地学习序列 :cite:`Hochreiter.Bengio.Frasconi.ea.2001` 中的远距离依赖关系。 ![Comparing CNN (padding tokens are omitted), RNN, and self-attention architectures.](../img/cnn-rnn-self-attention.svg) :label:`fig_cnn-rnn-self-attention` 考虑一个内核大小为 $k$ 的卷积层。我们将在后面的章节中提供有关使用 CNN 处理序列的更多详细信息目前,我们只需要知道,由于序列长度是 $n$,所以输入和输出通道的数量都是 $d$,卷积层的计算复杂度为 $\mathcal{O}(knd^2)$。如 :numref:`fig_cnn-rnn-self-attention` 所示,CNN 是分层的,因此有 $\mathcal{O}(1)$ 个顺序操作,最大路径长度为 $\mathcal{O}(n/k)$。例如,$\mathbf{x}_1$ 和 $\mathbf{x}_5$ 处于 :numref:`fig_cnn-rnn-self-attention` 中内核大小为 3 的双层 CNN 的接受范围内。 更新 rnN 的隐藏状态时,$d \times d$ 权重矩阵和 $d$ 维隐藏状态的乘法计算复杂度为 $\mathcal{O}(d^2)$。由于序列长度为 $n$,因此循环层的计算复杂度为 $\mathcal{O}(nd^2)$。根据 :numref:`fig_cnn-rnn-self-attention`,有 $\mathcal{O}(n)$ 个顺序操作无法并行化,最大路径长度也是 $\mathcal{O}(n)$。 在自我注意中,查询、键和值都是 $n \times d$ 矩阵。考虑 :eqref:`eq_softmax_QK_V` 中的扩展点-产品关注点,其中 $n \times d$ 矩阵乘以 $d \times n$ 矩阵,然后输出 $n \times n$ 矩阵乘以 $n \times d$ 矩阵。因此,自我注意力具有 $\mathcal{O}(n^2d)$ 计算复杂性。正如我们在 :numref:`fig_cnn-rnn-self-attention` 中看到的那样,每个令牌都通过自我注意直接连接到任何其他令牌。因此,计算可以与 $\mathcal{O}(1)$ 顺序操作并行,最大路径长度也是 $\mathcal{O}(1)$。 总而言之,CNN 和自我注意力都可以享受并行计算,而且自我注意力的最大路径长度最短。但是,相对于序列长度的二次计算复杂性使得自我注意力在很长的序列中非常缓慢。 ## 位置编码 :label:`subsec_positional-encoding` 与逐个重复处理序列令牌的 RNN 不同,自我注意力会放弃顺序操作,而倾向于并行计算。要使用序列顺序信息,我们可以通过在输入表示中添加 * 位置编码 * 来注入绝对或相对位置信息。可以学习或修复位置编码。在下面,我们描述了基于正弦和余弦函数 :cite:`Vaswani.Shazeer.Parmar.ea.2017` 的固定位置编码。 假设输入表示 $\mathbf{X} \in \mathbb{R}^{n \times d}$ 包含一个序列中 $n$ 令牌的 $d$ 维嵌入。位置编码使用相同形状的位置嵌入矩阵 $\mathbf{P} \in \mathbb{R}^{n \times d}$ 输出 $\mathbf{X} + \mathbf{P}$,该矩阵在 $i^\mathrm{th}$ 行和 $(2j)^\mathrm{th}$ 或 $(2j + 1)^\mathrm{th}$ 列上的元素为 $$\begin{aligned} p_{i, 2j} &= \sin\left(\frac{i}{10000^{2j/d}}\right),\\p_{i, 2j+1} &= \cos\left(\frac{i}{10000^{2j/d}}\right).\end{aligned}$$ :eqlabel:`eq_positional-encoding-def` 乍一看,这种三角函数设计看起来很奇怪。在解释这个设计之前,让我们首先在下面的 `PositionalEncoding` 课中实现它。 <jupyter_code>#@save class PositionalEncoding(nn.Module): def __init__(self, num_hiddens, dropout, max_len=1000): super(PositionalEncoding, self).__init__() self.dropout = nn.Dropout(dropout) # Create a long enough `P` self.P = torch.zeros((1, max_len, num_hiddens)) X = torch.arange(max_len, dtype=torch.float32).reshape( -1, 1) / torch.pow( 10000, torch.arange(0, num_hiddens, 2, dtype=torch.float32) / num_hiddens) self.P[:, :, 0::2] = torch.sin(X) self.P[:, :, 1::2] = torch.cos(X) def forward(self, X): X = X + self.P[:, :X.shape[1], :].to(X.device) return self.dropout(X)<jupyter_output><empty_output><jupyter_text>在位置嵌入矩阵 $\mathbf{P}$ 中,行对应于序列中的位置,列表示不同的位置编码维度。在下面的示例中,我们可以看到位置嵌入矩阵的 $6^{\mathrm{th}}$ 和 $7^{\mathrm{th}}$ 列的频率高于 $8^{\mathrm{th}}$ 和 $9^{\mathrm{th}}$ 列。$6^{\mathrm{th}}$ 和 $7^{\mathrm{th}}$ 列之间的偏移量($8^{\mathrm{th}}$ 和 $9^{\mathrm{th}}$ 列相同)是由于正弦函数和余弦函数的交替。 <jupyter_code>encoding_dim, num_steps = 32, 60 pos_encoding = PositionalEncoding(encoding_dim, 0) pos_encoding.eval() X = pos_encoding(torch.zeros((1, num_steps, encoding_dim))) P = pos_encoding.P[:, :X.shape[1], :] d2l.plot(torch.arange(num_steps), P[0, :, 6:10].T, xlabel='Row (position)', figsize=(6, 2.5), legend=["Col %d" % d for d in torch.arange(6, 10)])<jupyter_output><empty_output><jupyter_text>### 绝对位置信息 要了解沿编码维度单调降低的频率与绝对位置信息的关系,让我们打印出 $0, 1, \ldots, 7$ 的二进制表示形式。正如我们所看到的,每个数字、每两个数字和每四个数字上的最低位、第二位和第三位最低位分别交替。 <jupyter_code>for i in range(8): print(f'{i} in binary is {i:>03b}')<jupyter_output>0 in binary is 000 1 in binary is 001 2 in binary is 010 3 in binary is 011 4 in binary is 100 5 in binary is 101 6 in binary is 110 7 in binary is 111 <jupyter_text>在二进制表示中,较高的位比特的频率低于较低的位。同样,如下面的热图所示,位置编码通过使用三角函数在编码维度下降频率。由于输出是浮点数,因此此类连续表示比二进制表示法更节省空间。 <jupyter_code>P = P[0, :, :].unsqueeze(0).unsqueeze(0) d2l.show_heatmaps(P, xlabel='Column (encoding dimension)', ylabel='Row (position)', figsize=(3.5, 4), cmap='Blues')<jupyter_output><empty_output>
no_license
/chapter_attention-mechanisms/self-attention-and-positional-encoding.ipynb
Global19/d2l-zh-pytorch-sagemaker
7
<jupyter_start><jupyter_text> # BERT 모델 --- ## KorQuAD 데이터셋 KorQuAD(The Korean Question Answering Dataset, 한국어 질의응답 데이터셋) 을 가지고 자연어처리(Natural Language Processing) 분야의 기계독해(Machine Reading Comprehension, MRC) 태스크 처리 KorQuAD 데이터셋은 스탠포드 대학에서 구축한 대용량 데이터셋 SQuAD 데이터셋을 벤치마크. SQuAD 데이터셋은 '기계독해(MRC)' 태스크 처리를 테스트 하기 위한 데이터 셋으로, 언어 모델 성능 측정의 가장 표준적인 벤치마크입니다. 자연어처리 분야에는 '감성분석(Sentimental Analysis)', '번역(Translation)' 등 여러 태스크가 있지만, 사람의 질문에 대해서 자연어의 의미를 정확히 이해하고 답할 수 있는지를 측정하는 '기계독해(MRC)' 태스크가 매우 중요합니다. 참고. [SQuAD 공식페이지](https://rajpurkar.github.io/SQuAD-explorer/) [KorQuAD 공식페이지](https://korquad.github.io/) [MRC 모델, 어떻게 개발하고 평가하나요?](https://blog.naver.com/skelterlabs/222025030327) ### KorQuAD1.0 과 KorQuAD2.0 주요 차이 1. 문서의 길이 KorQuAD 1.0 : 한 두 문단 정도의 지문 길이 KorQuAD 2.0 : 위키백과 한 페이지 전체 지문 길이 2. 문서의 구조 KorQuAD 1.0 : 단순 string 으로 구성된 텍스트 문서 (?) KorQuAD 2.0: 지문에 리스트와 표가 포함. MRC 모델이 html tag 로 구조화된 문서를 이해할 수 있어야 합니다. 3. 답변의 길이 및 구조 KorQuAD 1.0 : 단순 단어나 구의 문장 범위 (?) KorQuAD 2.0 : 단어나 구의 단위를 넘어서 문단, 표, 리스트 전체를 포괄하는 긴 영역 포함. KorQuAD1.0 데이터 다운로드 ``` $ wget https://korquad.github.io/dataset/KorQuAD_v1.0_train.json $ wget https://korquad.github.io/dataset/KorQuAD_v1.0_dev.json ``` model, vocab, text corpus 데이터 다운로드 ``` $ wget https://aiffelstaticprd.blob.core.windows.net/media/documents/ko_32000.model $ wget https://aiffelstaticprd.blob.core.windows.net/media/documents/ko_32000.vocab $ wget https://aiffelstaticprd.blob.core.windows.net/media/documents/bert_pretrain_32000.hdf5 $ wget https://aiffelstaticprd.blob.core.windows.net/media/documents/kowiki.txt.zip ``` <jupyter_code># imports from __future__ import absolute_import, division, print_function, unicode_literals import tensorflow as tf import tensorflow.keras.backend as K import tensorflow_addons as tfa import os import re import numpy as np import pandas as pd import pickle import random import collections import json from datetime import datetime import sentencepiece as spm from tqdm.notebook import tqdm import matplotlib.pyplot as plt import seaborn as sns from wordcloud import WordCloud # 랜덤시드 고정 random_seed = 1234 random.seed(random_seed) np.random.seed(random_seed) tf.random.set_seed(random_seed)<jupyter_output><empty_output><jupyter_text> ## 데이터 확인 ### json 포맷 데이터 1. 파이썬 json 모듈의 dumps() 메소드로 확인 2. json 포맷의 파일을 리스트로 읽어와 리스트의 길이와 첫 번째 아이템만 확인하는 함수 작성 <jupyter_code># 파이썬 json 모듈의 dumps 메소드 사용 # 내용이 길기 때문에 주석처리 # print(json.dumps(train_json["data"][0], indent=2, ensure_ascii=False)) # json 포맷의 파일을 간략히 확인하기 위한 함수 작성 def print_json_tree(data, indent=""): for key, value in data.items(): # list 형태의 item 일 경우 if type(value) == list: print(f'{indent}- {key}: [{len(value)}]') # 리스트의 길이 출력 print_json_tree(value[0], indent + " ") # 리스트의 첫 번째 데이터 출력 # list 형태의 item 아닐 경우 else: print(f'{indent}- {key}: {value}') # 데이터 출력 # 데이터 불러오기 data_dir = os.getenv('HOME')+'/aiffel/bert_qna/data' model_dir = os.getenv('HOME')+'/aiffel/bert_qna/models' # Train 데이터 확인 train_json_path = data_dir + '/KorQuAD_v1.0_train.json' with open(train_json_path) as f: train_json = json.load(f) print_json_tree(train_json) # Test 데이터 확인 dev_json_path = data_dir + '/KorQuAD_v1.0_dev.json' with open(dev_json_path) as f: dev_json = json.load(f) print_json_tree(dev_json)<jupyter_output>- version: KorQuAD_v1.0_dev - data: [140] - paragraphs: [2] - qas: [7] - answers: [1] - text: 1989년 2월 15일 - answer_start: 0 - id: 6548850-0-0 - question: 임종석이 여의도 농민 폭력 시위를 주도한 혐의로 지명수배 된 날은? - context: 1989년 2월 15일 여의도 농민 폭력 시위를 주도한 혐의(폭력행위등처벌에관한법률위반)으로 지명수배되었다. 1989년 3월 12일 서울지방검찰청 공안부는 임종석의 사전구속영장을 발부받았다. 같은 해 6월 30일 평양축전에 임수경을 대표로 파견하여 국가보안법위반 혐의가 추가되었다. 경찰은 12월 18일~20일 사이 서울 경희대학교에서 임종석이 성명 발표를 추진하고 있다는 첩보를 입수했고, 12월 18일 오전 7시 40분 경 가스총과 전자봉으로 무장한 특공조 및 대공과 직원 12명 등 22명의 사복 경찰을 승용차 8대에 나누어 경희대학교에 투입했다. 1989년 12월 18일 오전 8시 15분 경 서울청량리경찰서는 호위 학생 5명과 함께 경희대학교 학생회관 건물 계단을 내려오는 임종석을 발견, 검거해 구속을 집행했다. 임종석은 청량리경찰서에서 약 1시간 동안 조사를 받은 뒤 오전 9시 50분 경 서울 장안동의 서울지방경찰청 공안분실로 인계되었다. - title: 임종석 <jupyter_text> ## 데이터 전처리 1 : 띄어쓰기 단위 (어절) Tokenize 하여 정보관리 '기계독해(MRC)' 태스크 처리를 위한 데이터 전처리는 다른 자연어처리 태스크와 다른 점이 있습니다. - 감성분석이나 번역 태스크의 자연어 전처리 띄어쓰기 등을 기준으로 어휘(또는 단어)를 나누고, 어휘에 속하지 않는 불용어(,.!?)를 제거한 뒤, 이를 Embedding vectorization 하여 벡터 공간상의 데이터로 변환해주는 작업을 주로 하였습니다. - 기계독해 태스크의 자연어 전처리 띄어쓰기 단위로 어휘(또는 단어)를 나누고, 나누어진 영역별로 위치(순서)정보를 인코딩해 줍니다. 이를 통해 문장 내 어휘의 위치정보를 학습할 수 있습니다. ### 문장을 '어절' 별로 나누어 '문자(character)' 마다 tokenize (인코딩) 값 저장 띄어쓰기 단위로 token을 정리한 후 word token 영역별로 유니크한 숫자(어절 번호)를 부여하는 함수를 작성합니다. 이후에 __형태소에 따라__ 문자(charater) 혹은 subword 단위로 token이 분리되는 것에 대비해서 원래 데이터가 띄어쓰기 단위로 어떠했었는지 word token 영역별로 추가 정보를 관리하면 도움이 됩니다. 문자(character)별로 word_token 영역을 표시해 주는 char_to_word 리스트를 생성해 관리해 둡니다. 이 값은 현재 글자가 몇 번째 어절에 포함된 것이었는지를 말해 줍니다. <jupyter_code># whitespace def _is_whitespace(c): if c == " " or c == "\t" or c == "\r" or c == "\n" or ord(c) == 0x202F: return True return False # whitespace가 2개인 경우도 아래와 같이 동일하게 출력되도록 처리해야 합니다. string1 = '1839년 파우스트을 읽었다.' string2 = '1839년 파우스트을 읽었다.' string1[6:10], string2[7:11] # 글자별로 띄어쓰기 영역 정보 관리시 문제점 word_tokens = [] char_to_word = [] prev_is_whitespace = True # 첫번째 문장(string1)에 대해 띄어쓰기 영역 정보를 표시 for c in string1: if _is_whitespace(c): prev_is_whitespace = True else: if prev_is_whitespace: word_tokens.append(c) else: word_tokens[-1] += c prev_is_whitespace = False char_to_word.append(len(word_tokens) - 1) print(f'\'{c}\' : {word_tokens} : {char_to_word}') # 글자별로 띄어쓰기 영역 정보 관리시 문제점 word_tokens = [] char_to_word = [] prev_is_whitespace = True # 두번째 문장(string2)에 대해 띄어쓰기 영역 정보를 표시 for c in string2: if _is_whitespace(c): prev_is_whitespace = True else: if prev_is_whitespace: word_tokens.append(c) else: word_tokens[-1] += c prev_is_whitespace = False char_to_word.append(len(word_tokens) - 1) print(f'\'{c}\' : {word_tokens} : {char_to_word}')<jupyter_output>'1' : ['1'] : [0] '8' : ['18'] : [0, 0] '3' : ['183'] : [0, 0, 0] '9' : ['1839'] : [0, 0, 0, 0] '년' : ['1839년'] : [0, 0, 0, 0, 0] ' ' : ['1839년'] : [0, 0, 0, 0, 0, 0] ' ' : ['1839년'] : [0, 0, 0, 0, 0, 0, 0] '파' : ['1839년', '파'] : [0, 0, 0, 0, 0, 0, 0, 1] '우' : ['1839년', '파우'] : [0, 0, 0, 0, 0, 0, 0, 1, 1] '스' : ['1839년', '파우스'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1] '트' : ['1839년', '파우스트'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1] '을' : ['1839년', '파우스트을'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1] ' ' : ['1839년', '파우스트을'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1] '읽' : ['1839년', '파우스트을', '읽'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2] '었' : ['1839년', '파우스트을', '읽었'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2] '다' : ['1839년', '파우스트을', '읽었다'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2] '.' : ['1839년', '파우스트을', '읽었다.'] : [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2] <jupyter_text> 공백 길이에 따라 두 문장의 영역표시 결과가 조금 달라집니다. 이러한 문제를 해결하기 위해 위에서 정의한 ```_is_whitespace()``` 함수를 활용하여 띄어쓰기 단위로 어휘를 나누는 인코딩(?) 함수를 작성해 줍니다. <jupyter_code># 띄어쓰기 단위로 어휘의 위치정보를 나누는 인코딩 함수 작성 # _is_whitespace() 적용 def _tokenize_whitespace(string): word_tokens = [] char_to_word = [] prev_is_whitespace = True for c in string: if _is_whitespace(c): prev_is_whitespace = True else: if prev_is_whitespace: word_tokens.append(c) else: word_tokens[-1] += c prev_is_whitespace = False char_to_word.append(len(word_tokens) - 1) return word_tokens, char_to_word # 첫번째 문장(string1)에 대해 띄어쓰기 영역 정보를 표시 word_tokens, char_to_word = _tokenize_whitespace(string1) for c, i in zip(list(string1), char_to_word): print(f'\'{c}\' : {i}') word_tokens, char_to_word # 두번째 문장(string2)에 대해 띄어쓰기 영역 정보를 표시 word_tokens, char_to_word = _tokenize_whitespace(string2) for c, i in zip(list(string2), char_to_word): print(f'\'{c}\' : {i}') word_tokens, char_to_word<jupyter_output>'1' : 0 '8' : 0 '3' : 0 '9' : 0 '년' : 0 ' ' : 0 ' ' : 0 '파' : 1 '우' : 1 '스' : 1 '트' : 1 '을' : 1 ' ' : 1 '읽' : 2 '었' : 2 '다' : 2 '.' : 2 <jupyter_text> ## 데이터 전처리 2 : 단어사전(Vocab)으로 Tokenize ### Subword Segmentation (형태소 분할) 어절을 또 다시 '[__형태소__](https://terms.naver.com/entry.naver?docId=3559671&cid=58583&categoryId=58729)' 단위로 나누어 Tokenize 합니다. '읽다', '읽었다', '읽으려고', '읽고' 등의 많은 데이터를 의미 형태소 '읽' 과 의존 형태소 '다' 등으로 나누어 저장합니다. ### SentencePiece 모델 Vocab BERT에는 WordPiece 모델 사용이 일반적이지만, 여기에서는 SentencePiece 모델을 이용해 Subword(형태소) 기반의 텍스트 전처리를 진행합니다. SentencePiece 모델은 구글에서 오픈소스로 제공하는 형태소 분할 모델입니다. 파이썬에서 손쉽게 사용가능하며, WordPiece 등 다른 모델들을 통합하여 제공하므로 최근 널리 사용되고 있습니다. SentencePiece 같은 모델들은 언어마다 다른 문법규칙을 활용하지 않고 어떤 언어에든 보편적으로 적용 가능하다는 장점이 있습니다. - 적절한 Subword(형태소) 분절 규칙을 학습하거나, - 자주 사용되는 구문을 하나의 단어로 묶어내는 등의 통계적인 방법을 사용합니다. <jupyter_code># sentencepiece 모듈을 이용한 형태소 단위 분할 확인 # vocab 모델 loading vocab = spm.SentencePieceProcessor() vocab.load(f"{model_dir}/ko_32000.model") # word(어절)를 subword(형태소)로 변경하면서 index 저장 word_to_token = [] context_tokens = [] for (i, word) in enumerate(word_tokens): word_to_token.append(len(context_tokens)) # 형태소로 쪼개지기 전의 어절 단위의 시작지점 토큰 정보 저장 tokens = vocab.encode_as_pieces(word) # SentencePiece 를 사용해 Subword(형태소) 로 분할 for token in tokens: context_tokens.append(token) context_tokens, word_to_token<jupyter_output><empty_output><jupyter_text> ### 띄어쓰기에 따른 '어절 정보' 압축 저장 위 전처리 1 단계 '어절 정보에 따른 tokenize' 단계에서 우리는 tokenize 된 정보를 '문자(character)' 단위로 인코딩해 두었습니다. 이를 이용하여 전처리 2 단계 '형태소 정보에 따른 tokenize' 단계에서 어절 단위에 따른 띄어쓰기 정보가 소실되지 않도록, 형태소에 따라 tokenize 된 데이터에 띄어쓰기 위치 정보를 함께 저장합니다. word_to_token의 \[0, 2, 5\] 는 context_tokens 에 쪼개져 담긴 0번, 2번, 5번 토큰인 '▁1839', '▁', '▁읽' 이 어절단위의 첫번째 토큰이 된다는 (띄어쓰기) 정보를 담아둔 것입니다. <jupyter_code># sentencepiece 모듈을 활용한 형태소 분할 함수 작성 def _tokenize_vocab(vocab, context_words): word_to_token = [] context_tokens = [] for (i, word) in enumerate(context_words): word_to_token.append(len(context_tokens)) tokens = vocab.encode_as_pieces(word) for token in tokens: context_tokens.append(token) return context_tokens, word_to_token print(word_tokens) # 처리해야 할 word 단위 입력 context_tokens, word_to_token = _tokenize_vocab(vocab, word_tokens) context_tokens, word_to_token # Subword 단위로 토큰화한 결과<jupyter_output>['1839년', '파우스트을', '읽었다.'] <jupyter_text> ## 데이터 전처리 3 : Improve Span KorQuAD 데이터셋은 - 질문(question)과 지문(context)을 주고, - 지문 영역에서 정답(answer)을 찾도록 구성 그러므로 정답에 해당하는 지문 영역을 정확히 찾아내는 것이 전처리의 핵심적인 작업이 됩니다. KorQuAD 데이터셋에서 context, question, answer 를 추출해 봅시다. <jupyter_code># KorQuAD 데이터셋에서 context, question, answer 정보 추출 context = train_json['data'][0]['paragraphs'][0]['context'] question = train_json['data'][0]['paragraphs'][0]['qas'][0]['question'] answer_text = train_json['data'][0]['paragraphs'][0]['qas'][0]['answers'][0]['text'] answer_start = train_json['data'][0]['paragraphs'][0]['qas'][0]['answers'][0]['answer_start'] answer_end = answer_start + len(answer_text) - 1 print('[context] ', context) print('[question] ', question) print('[answer] ', answer_text, '\n') # context 에 포함된 answer 의 문자단위 시작 인덱스와 종료 인덱스 저장 ! print('[answer_start] index: ', answer_start, 'character: ', context[answer_start]) print('[answer_end]index: ', answer_end, 'character: ', context[answer_end]) # answer_text에 해당하는 context 영역을 정확히 찾아내야 합니다. assert context[answer_start:answer_end + 1] == answer_text<jupyter_output>[context] 1839년 바그너는 괴테의 파우스트을 처음 읽고 그 내용에 마음이 끌려 이를 소재로 해서 하나의 교향곡을 쓰려는 뜻을 갖는다. 이 시기 바그너는 1838년에 빛 독촉으로 산전수전을 다 걲은 상황이라 좌절과 실망에 가득했으며 메피스토펠레스를 만나는 파우스트의 심경에 공감했다고 한다. 또한 파리에서 아브네크의 지휘로 파리 음악원 관현악단이 연주하는 베토벤의 교향곡 9번을 듣고 깊은 감명을 받았는데, 이것이 이듬해 1월에 파우스트의 서곡으로 쓰여진 이 작품에 조금이라도 영향을 끼쳤으리라는 것은 의심할 여지가 없다. 여기의 라단조 조성의 경우에도 그의 전기에 적혀 있는 것처럼 단순한 정신적 피로나 실의가 반영된 것이 아니라 베토벤의 합창교향곡 조성의 영향을 받은 것을 볼 수 있다. 그렇게 교향곡 작곡을 1839년부터 40년에 걸쳐 파리에서 착수했으나 1악장을 쓴 뒤에 중단했다. 또한 작품의 완성과 동시에 그는 이 서곡(1악장)을 파리 음악원의 연주회에서 연주할 파트보까지 준비하였으나, 실제로는 이루어지지는 않았다. 결국 초연은 4년 반이 지난 후에 드레스덴에서 연주되었고 재연도 이루어졌지만, 이후에 그대로 방치되고 말았다. 그 사이에 그는 리엔치와 방황하는 네덜란드인을 완성하고 탄호이저에도 착수하는 등 분주한 시간을 보냈는데, 그런 바쁜 생활이 이 곡을 잊게 한 것이 아닌가 하는 의견도 있다. [question] 바그너는 괴테의 파우스트를 읽고 무엇을 쓰고자 했는가? [answer] 교향곡 [answer_start] index: 54 character: 교 [answer_end]index: 56 character: 곡 <jupyter_text> ### context 데이터 전처리 1, 2 단계 처리 확인 <jupyter_code># context 를 띄어쓰기 (어절,word) 단위로 토큰화한 결과 확인 # 위에서 정의한 _tokenize_whitespace() 함수 사용 word_tokens, char_to_word = _tokenize_whitespace(context) print( word_tokens[:20]) char_to_word[:20], context[:20] # 띄어쓰기 (어절,word) 단위로 쪼개진 context(word_tokens) 를 형태소(Subword)로 토큰화한 결과 확인 # 위에서 정의한 _tokenize_vocab() 함수 사용 context_tokens, word_to_token = _tokenize_vocab(vocab, word_tokens) for i in range(min(20, len(word_to_token) - 1)): print(word_to_token[i], context_tokens[word_to_token[i]:word_to_token[i + 1]])<jupyter_output>0 ['▁1839', '년'] 2 ['▁바그너', '는'] 4 ['▁괴테', '의'] 6 ['▁', '파우스트', '을'] 9 ['▁처음'] 10 ['▁읽고'] 11 ['▁그'] 12 ['▁내용에'] 13 ['▁마음이'] 14 ['▁끌려'] 15 ['▁이를'] 16 ['▁소재로'] 17 ['▁해서'] 18 ['▁하나의'] 19 ['▁교향곡', '을'] 21 ['▁쓰', '려는'] 23 ['▁뜻을'] 24 ['▁갖는다', '.'] 26 ['▁이'] 27 ['▁시기'] <jupyter_text> ### answer 데이터 전처리 1, 2 단계 처리 확인 위에서 우리는 context에 포함된 answer의 글자단위 시작 인덱스(answer_start)와 종료 인덱스(answer_end)를 구했습니다. 이 위치를 어절(word) 단위로 변환하면 어떻게 될까요? <jupyter_code># answer_start 와 answer_end 로부터 word_start 와 word_end 를 생성 word_start = char_to_word[answer_start] word_end = char_to_word[answer_end] word_start, word_end, answer_text, word_tokens[word_start:word_end + 1]<jupyter_output><empty_output><jupyter_text> 우리가 찾는 정답은 15번째 어절(index=14)에 있었군요. 하지만 우리가 원하는 정답은 '교향곡'이지, '교향곡을'은 아닙니다. 그래서 이번에는 word_start로부터 word_end까지의 context를 Subword 단위로 토큰화한 결과를 살펴봅시다. <jupyter_code># token_start = word_to_token[word_start] if word_end < len(word_to_token) - 1: token_end = word_to_token[word_end + 1] - 1 else: token_end = len(context_tokens) - 1 token_start, token_end, context_tokens[token_start:token_end + 1]<jupyter_output><empty_output><jupyter_text> 거의 정답에 근접했습니다. Subword 단위로 토큰화한 결과 중에는 우리가 찾는 정답과 정확히 일치하는 답이 있는것 같습니다. <jupyter_code># 실제 정답인 answer_text 도 Subword 기준으로 토큰화 token_answer = " ".join(vocab.encode_as_pieces(answer_text)) token_answer<jupyter_output><empty_output><jupyter_text> 이제 눈으로 봐도 어디가 정확히 정답인지 알 수 있게 되었지만, 좀더 일반적인 방법으로 정답 토큰 범위를 찾는 코드를 작성해 보겠습니다. KorQuAD 문제의 정답은 이번처럼 단답형만 있는 것은 아니기 때문입니다. <jupyter_code># 정답이 될수 있는 new_start 와 new_end 의 경우를 순회탐색하여 확인 for new_start in range(token_start, token_end + 1): for new_end in range(token_end, new_start - 1, -1): text_span = " ".join(context_tokens[new_start : (new_end + 1)]) if text_span == token_answer: # 정답과 일치하는 경우 print("O >>", (new_start, new_end), text_span) else: print("X >>", (new_start, new_end), text_span)<jupyter_output>X >> (19, 20) ▁교향곡 을 O >> (19, 19) ▁교향곡 X >> (20, 20) 을 <jupyter_text> context에서 answer의 위치를 토큰화된 상태에서 찾는 함수를 아래와 같이 작성 <jupyter_code># context_tokens에서 char_answer의 위치를 찾아 리턴하는 함수 작성 def _improve_span(vocab, context_tokens, token_start, token_end, char_answer): token_answer = " ".join(vocab.encode_as_pieces(char_answer)) for new_start in range(token_start, token_end + 1): for new_end in range(token_end, new_start - 1, -1): text_span = " ".join(context_tokens[new_start : (new_end + 1)]) if text_span == token_answer: return (new_start, new_end) return (token_start, token_end) # 확인 token_start, token_end = _improve_span(vocab, context_tokens, token_start, token_end, answer_text) print('token_start:', token_start, ' token_end:', token_end) context_tokens[token_start:token_end + 1]<jupyter_output>token_start: 19 token_end: 19 <jupyter_text> ## 데이터 전처리 4 : 데이터셋 분리 train 데이터셋, dev 데이터셋을 분리하여, 위에서 작성한 ```_improve_span()``` 함수를 이용해 전처리 후 파일로 저장합니다. <jupyter_code># 데이터셋 분리하는 함수 작성 def dump_korquad(vocab, json_data, out_file): with open(out_file, "w") as f: for data in tqdm(json_data["data"]): title = data["title"] for paragraph in data["paragraphs"]: context = paragraph["context"] context_words, char_to_word = _tokenize_whitespace(context) for qa in paragraph["qas"]: assert len(qa["answers"]) == 1 qa_id = qa["id"] question = qa["question"] answer_text = qa["answers"][0]["text"] answer_start = qa["answers"][0]["answer_start"] answer_end = answer_start + len(answer_text) - 1 assert answer_text == context[answer_start:answer_end + 1] word_start = char_to_word[answer_start] word_end = char_to_word[answer_end] word_answer = " ".join(context_words[word_start:word_end + 1]) char_answer = " ".join(answer_text.strip().split()) assert char_answer in word_answer context_tokens, word_to_token = _tokenize_vocab(vocab, context_words) token_start = word_to_token[word_start] if word_end < len(word_to_token) - 1: token_end = word_to_token[word_end + 1] - 1 else: token_end = len(context_tokens) - 1 token_start, token_end = _improve_span(vocab, context_tokens, token_start, token_end, char_answer) data = {"qa_id": qa_id, "title": title, "question": vocab.encode_as_pieces(question), "context": context_tokens, "answer": char_answer, "token_start": token_start, "token_end":token_end} f.write(json.dumps(data, ensure_ascii=False)) f.write("\n") # 전처리를 수행하여 파일로 생성 dump_korquad(vocab, train_json, f"{data_dir}/korquad_train.json") dump_korquad(vocab, dev_json, f"{data_dir}/korquad_dev.json") # 전처리가 잘 되었는지 실제로 파일 내용을 확인 def print_file(filename, count=10): """ 파일 내용 출력 :param filename: 파일 이름 :param count: 출력 라인 수 """ with open(filename) as f: for i, line in enumerate(f): if count <= i: break print(line.strip()) print_file(f"{data_dir}/korquad_train.json")<jupyter_output>{"qa_id": "6566495-0-0", "title": "파우스트_서곡", "question": ["▁바그너", "는", "▁괴테", "의", "▁", "파우스트", "를", "▁읽고", "▁무엇을", "▁쓰고", "자", "▁", "했", "는", "가", "?"], "context": ["▁1839", "년", "▁바그너", "는", "▁괴테", "의", "▁", "파우스트", "을", "▁처음", "▁읽고", "▁그", "▁내용에", "▁마음이", "▁끌려", "▁이를", "▁소재로", "▁해서", "▁하나의", "▁교향곡", "을", "▁쓰", "려는", "▁뜻을", "▁갖는다", ".", "▁이", "▁시기", "▁바그너", "는", "▁1838", "년에", "▁빛", "▁독", "촉", "으로", "▁산", "전", "수", "전을", "▁다", "▁", "걲", "은", "▁상황이", "라", "▁좌절", "과", "▁실망", "에", "▁가득", "했으며", "▁메", "피스", "토", "펠", "레스", "를", "▁만나는", "▁", "파우스트", "의", "▁심", "경에", "▁공감", "했다고", "▁한다", ".", "▁또한", "▁파리에서", "▁아브", "네", "크의", "▁지휘", "로", "▁파리", "▁음악원", "▁관현악단", "이", "▁연주하는", "▁베토벤", "의", "▁교향곡", "▁9", "번을", "▁듣고", "▁깊은", "▁감", "명을", "▁받았는데", ",", "▁이것이", "▁이듬해", "▁1", "월에", "▁", "파우스트", "의", "▁서", "곡으로", "▁쓰여진", "▁이", "▁작품에", "▁조금", "이라도", "▁영향을", "▁끼", "쳤", "으리라", "는", "▁것은", "▁의심", "할", "▁여지가", "▁없다", ".", "▁여기", "의", "▁라", "단", "조", "▁조성", "의", "▁경우에도", "▁그의", "▁전기", "에", "▁적혀", "▁있는", [...]<jupyter_text> ## 데이터 전처리 5 : 데이터 분석 원본 데이터셋을 전처리하여 우리의 모델이 다루게 될 데이터셋으로 가공하는 과정을 진행하였습니다. 그러나 이 데이터셋을 그대로 사용할 수 있을지, 혹은 이상(abnormal) 데이터가 존재하지는 않는지 분석하는 과정이 필요합니다. <jupyter_code># 데이터 분석을 위해 전처리된 데이터를 question, context, answer 별로 불러오기 questions = [] contexts = [] token_starts = [] with open(f"{data_dir}/korquad_train.json") as f: for i, line in enumerate(f): data = json.loads(line) questions.append(data["question"]) contexts.append(data["context"]) token_starts.append(data["token_start"]) if i < 10: print(data["token_start"], data["question"])<jupyter_output>19 ['▁바그너', '는', '▁괴테', '의', '▁', '파우스트', '를', '▁읽고', '▁무엇을', '▁쓰고', '자', '▁', '했', '는', '가', '?'] 168 ['▁바그너', '는', '▁교향곡', '▁작곡', '을', '▁어디', '까지', '▁쓴', '▁뒤에', '▁중단', '했', '는', '가', '?'] 80 ['▁바그너', '가', '▁', '파우스트', '▁서', '곡을', '▁쓸', '▁때', '▁어떤', '▁곡', '의', '▁영향을', '▁받았', '는', '가', '?'] 6 ['▁1839', '년', '▁바그너', '가', '▁교향곡', '의', '▁소재로', '▁쓰', '려고', '▁했던', '▁책은', '?'] 143 ['▁', '파우스트', '▁서', '곡', '의', '▁라', '단', '조', '▁조성', '이', '▁영향을', '▁받은', '▁베토벤', '의', '▁곡은', '?'] 0 ['▁바그너', '가', '▁', '파우스트', '를', '▁처음으로', '▁읽', '은', '▁', '년', '도', '는', '?'] 165 ['▁바그너', '가', '▁처음', '▁교향곡', '▁작곡', '을', '▁한', '▁장소', '는', '?'] 216 ['▁바그너', '의', '▁1', '악장', '의', '▁초연', '은', '▁어디서', '▁연주', '되었', '는', '가', '?'] 164 ['▁바그너', '의', '▁작품을', '▁시인', '의', '▁피로', '▁쓰여', '졌다', '고', '▁극찬', '한', '▁것은', '▁누구', '인', '가', '?'] 7 ['▁잊', '혀', '져', '▁있는', '▁', '파우스트', '▁서', '곡', '▁1', '악장', '을', '▁부활', '시킨', '▁것은', '▁누구', '인', '가', '?'] <jupyter_text> ### 전체 데이터에서 question 항목의 길이 분포 확인 <jupyter_code># token count train_question_counts = [len(question) for question in questions] train_question_counts[:10] # 그래프에 대한 이미지 사이즈 선언 # figsize: (가로, 세로) 형태의 튜플로 입력 plt.figure(figsize=(8, 4)) # histogram 선언 # bins: 히스토그램 값들에 대한 버켓 범위, # range: x축 값의 범위 # facecolor: 그래프 색상 # label: 그래프에 대한 라벨 plt.hist(train_question_counts, bins=100, range=[0, 100], facecolor='b', label='train') # 그래프 제목 plt.title('Count of question') # 그래프 x 축 라벨 plt.xlabel('Number of question') # 그래프 y 축 라벨 plt.ylabel('Count of question') plt.show() # 데이터 길이 print(f"question 길이 최대: {np.max(train_question_counts):4d}") print(f"question 길이 최소: {np.min(train_question_counts):4d}") print(f"question 길이 평균: {np.mean(train_question_counts):7.2f}") print(f"question 길이 표준편차: {np.std(train_question_counts):7.2f}") # https://ko.wikipedia.org/wiki/%EB%B0%B1%EB%B6%84%EC%9C%84%EC%88%98 # 백분위수(Percentile)는 크기가 있는 값들로 이뤄진 자료를 순서대로 나열했을 때 백분율로 나타낸 특정 위치의 값을 이르는 용어이다. # 일반적으로 크기가 작은 것부터 나열하여 가장 작은 것을 0, 가장 큰 것을 100으로 한다. # 100개의 값을 가진 어떤 자료의 20 백분위수는 그 자료의 값들 중 20번째로 작은 값을 뜻한다. 50 백분위수는 중앙값과 같다. percentile25 = np.percentile(train_question_counts, 25) percentile50 = np.percentile(train_question_counts, 50) percentile75 = np.percentile(train_question_counts, 75) percentileIQR = percentile75 - percentile25 percentileMAX = percentile75 + percentileIQR * 1.5 print(f"question 25/100분위: {percentile25:7.2f}") print(f"question 50/100분위: {percentile50:7.2f}") print(f"question 75/100분위: {percentile75:7.2f}") print(f"question IQR: {percentileIQR:7.2f}") print(f"question MAX/100분위: {percentileMAX:7.2f}") plt.figure(figsize=(4, 6)) # 박스플롯 생성 # 첫번째 파라메터: 여러 분포에 대한 데이터 리스트를 # labels: 입력한 데이터에 대한 라벨 # showmeans: 평균값을 표현 # 참고: https://leebaro.tistory.com/entry/%EB%B0%95%EC%8A%A4-%ED%94%8C%EB%A1%AFbox-plot-%EC%84%A4%EB%AA%85 plt.boxplot(train_question_counts, labels=['token counts'], showmeans=True) plt.show()<jupyter_output><empty_output><jupyter_text> ### 전체 데이터에서 context 항목의 길이 분포 확인 <jupyter_code># token count train_context_counts = [len(context) for context in contexts] train_context_counts[:10] # 그래프에 대한 이미지 사이즈 선언 # figsize: (가로, 세로) 형태의 튜플로 입력 plt.figure(figsize=(8, 4)) # histogram 선언 # bins: 히스토그램 값들에 대한 버켓 범위, # range: x축 값의 범위 # facecolor: 그래프 색상 # label: 그래프에 대한 라벨 plt.hist(train_context_counts, bins=900, range=[100, 1000], facecolor='r', label='train') # 그래프 제목 plt.title('Count of context') # 그래프 x 축 라벨 plt.xlabel('Number of context') # 그래프 y 축 라벨 plt.ylabel('Count of context') plt.show() # 데이터 길이 print(f"context 길이 최대: {np.max(train_context_counts):4d}") print(f"context 길이 최소: {np.min(train_context_counts):4d}") print(f"context 길이 평균: {np.mean(train_context_counts):7.2f}") print(f"context 길이 표준편차: {np.std(train_context_counts):7.2f}") # https://ko.wikipedia.org/wiki/%EB%B0%B1%EB%B6%84%EC%9C%84%EC%88%98 # 백분위수(Percentile)는 크기가 있는 값들로 이뤄진 자료를 순서대로 나열했을 때 백분율로 나타낸 특정 위치의 값을 이르는 용어이다. # 일반적으로 크기가 작은 것부터 나열하여 가장 작은 것을 0, 가장 큰 것을 100으로 한다. # 100개의 값을 가진 어떤 자료의 20 백분위수는 그 자료의 값들 중 20번째로 작은 값을 뜻한다. 50 백분위수는 중앙값과 같다. percentile25 = np.percentile(train_context_counts, 25) percentile50 = np.percentile(train_context_counts, 50) percentile75 = np.percentile(train_context_counts, 75) percentileIQR = percentile75 - percentile25 percentileMAX = percentile75 + percentileIQR * 1.5 print(f"context 25/100분위: {percentile25:7.2f}") print(f"context 50/100분위: {percentile50:7.2f}") print(f"context 75/100분위: {percentile75:7.2f}") print(f"context IQR: {percentileIQR:7.2f}") print(f"context MAX/100분위: {percentileMAX:7.2f}") plt.figure(figsize=(4, 6)) # 박스플롯 생성 # 첫번째 파라메터: 여러 분포에 대한 데이터 리스트를 # labels: 입력한 데이터에 대한 라벨 # showmeans: 평균값을 표현 # 참고: https://leebaro.tistory.com/entry/%EB%B0%95%EC%8A%A4-%ED%94%8C%EB%A1%AFbox-plot-%EC%84%A4%EB%AA%85 plt.boxplot(train_context_counts, labels=['token counts'], showmeans=True) plt.show()<jupyter_output><empty_output><jupyter_text> ### 전체 데이터에서 answer 항목의 길이 분포 확인 <jupyter_code># token count train_answer_starts = token_starts train_answer_starts[:10] # 그래프에 대한 이미지 사이즈 선언 # figsize: (가로, 세로) 형태의 튜플로 입력 plt.figure(figsize=(8, 4)) # histogram 선언 # bins: 히스토그램 값들에 대한 버켓 범위, # range: x축 값의 범위 # facecolor: 그래프 색상 # label: 그래프에 대한 라벨 plt.hist(train_answer_starts, bins=500, range=[0, 500], facecolor='g', label='train') # 그래프 제목 plt.title('Count of answer') # 그래프 x 축 라벨 plt.xlabel('Number of answer') # 그래프 y 축 라벨 plt.ylabel('Count of answer') plt.show() # 데이터 길이 print(f"answer 위치 최대: {np.max(train_answer_starts):4d}") print(f"answer 위치 최소: {np.min(train_answer_starts):4d}") print(f"answer 위치 평균: {np.mean(train_answer_starts):7.2f}") print(f"answer 위치 표준편차: {np.std(train_answer_starts):7.2f}") # https://ko.wikipedia.org/wiki/%EB%B0%B1%EB%B6%84%EC%9C%84%EC%88%98 # 백분위수(Percentile)는 크기가 있는 값들로 이뤄진 자료를 순서대로 나열했을 때 백분율로 나타낸 특정 위치의 값을 이르는 용어이다. # 일반적으로 크기가 작은 것부터 나열하여 가장 작은 것을 0, 가장 큰 것을 100으로 한다. # 100개의 값을 가진 어떤 자료의 20 백분위수는 그 자료의 값들 중 20번째로 작은 값을 뜻한다. 50 백분위수는 중앙값과 같다. percentile25 = np.percentile(train_answer_starts, 25) percentile50 = np.percentile(train_answer_starts, 50) percentile75 = np.percentile(train_answer_starts, 75) percentileIQR = percentile75 - percentile25 percentileMAX = percentile75 + percentileIQR * 1.5 print(f"answer 25/100분위: {percentile25:7.2f}") print(f"answer 50/100분위: {percentile50:7.2f}") print(f"answer 75/100분위: {percentile75:7.2f}") print(f"answer IQR: {percentileIQR:7.2f}") print(f"answer MAX/100분위: {percentileMAX:7.2f}") plt.figure(figsize=(4, 6)) # 박스플롯 생성 # 첫번째 파라메터: 여러 분포에 대한 데이터 리스트를 # labels: 입력한 데이터에 대한 라벨 # showmeans: 평균값을 표현 # 참고: https://leebaro.tistory.com/entry/%EB%B0%95%EC%8A%A4-%ED%94%8C%EB%A1%AFbox-plot-%EC%84%A4%EB%AA%85 plt.boxplot(train_answer_starts, labels=['token counts'], showmeans=True) plt.show()<jupyter_output><empty_output><jupyter_text> ## 데이터 전처리 5 : Word Cloud (자료 빈도수 시각화) 워드 클라우드(Word Cloud)란 자료의 빈도수를 시각화해서 나타내는 방법입니다. 문서의 핵심 단어를 한눈에 파악할 수 있고, 빅데이터를 분석할 때 데이터의 특징을 도출하기 위해서 활용됩니다. 빈도수가 높은 단어일수록 글씨 크기가 큰 특징이 있습니다. <jupyter_code># train documents documents = [] # 전체 데이터에서 title, context, question 문장을 모두 추출합니다. for data in tqdm(train_json["data"]): title = data["title"] documents.append(title) for paragraph in data["paragraphs"]: context = paragraph["context"] documents.append(context) for qa in paragraph["qas"]: assert len(qa["answers"]) == 1 question = qa["question"] documents.append(question) documents[:10] # 그중 맨 앞 10개만 확인해 봅니다. # documents를 전부 이어 하나의 문장으로 만들면 이렇게 보입니다. " ".join(documents[:10]) # WordCloud로 " ".join(documents)를 처리해 봅니다. wordcloud = WordCloud(width=800, height=800, font_path='/usr/share/fonts/truetype/nanum/NanumBarunGothic.ttf').generate(" ".join(documents)) plt.figure(figsize=(10, 10)) # image 출력, interpolation 이미지 시각화 옵션 plt.imshow(wordcloud, interpolation='bilinear') plt.axis('off') plt.show()<jupyter_output><empty_output><jupyter_text> ## 데이터 전처리 5 : 데이터 로드 지금까지 만든 데이터셋을 메모리에 로드합니다. <jupyter_code>train_json = os.path.join(data_dir, "korquad_train.json") dev_json = os.path.join(data_dir, "korquad_dev.json") class Config(dict): """ json을 config 형태로 사용하기 위한 Class :param dict: config dictionary """ __getattr__ = dict.__getitem__ __setattr__ = dict.__setitem__ args = Config({ 'max_seq_length': 384, 'max_query_length': 64, }) args # 생성한 데이터셋 파일을 메모리에 로딩하는 함수 def load_data(args, filename): inputs, segments, labels_start, labels_end = [], [], [], [] n_discard = 0 with open(filename, "r") as f: for i, line in enumerate(tqdm(f, desc=f"Loading ...")): data = json.loads(line) token_start = data.get("token_start") token_end = data.get("token_end") question = data["question"][:args.max_query_length] context = data["context"] answer_tokens = " ".join(context[token_start:token_end + 1]) context_len = args.max_seq_length - len(question) - 3 if token_end >= context_len: # 최대 길이내에 token이 들어가지 않은 경우 처리하지 않음 n_discard += 1 continue context = context[:context_len] assert len(question) + len(context) <= args.max_seq_length - 3 tokens = ['[CLS]'] + question + ['[SEP]'] + context + ['[SEP]'] ids = [vocab.piece_to_id(token) for token in tokens] ids += [0] * (args.max_seq_length - len(ids)) inputs.append(ids) segs = [0] * (len(question) + 2) + [1] * (len(context) + 1) segs += [0] * (args.max_seq_length - len(segs)) segments.append(segs) token_start += (len(question) + 2) labels_start.append(token_start) token_end += (len(question) + 2) labels_end.append(token_end) print(f'n_discard: {n_discard}') return (np.array(inputs), np.array(segments)), (np.array(labels_start), np.array(labels_end)) # train data load train_inputs, train_labels = load_data(args, train_json) print(f"train_inputs: {train_inputs[0].shape}") print(f"train_inputs: {train_inputs[1].shape}") print(f"train_labels: {train_labels[0].shape}") print(f"train_labels: {train_labels[1].shape}") # dev data load dev_inputs, dev_labels = load_data(args, dev_json) print(f"dev_inputs: {dev_inputs[0].shape}") print(f"dev_inputs: {dev_inputs[1].shape}") print(f"dev_labels: {dev_labels[0].shape}") print(f"dev_labels: {dev_labels[1].shape}") train_inputs[:10], train_labels[:10]<jupyter_output><empty_output><jupyter_text> ### 우리가 만든 데이터셋 확인 <jupyter_code># Question과 Context가 포함된 입력데이터 1번째 train_inputs[0][0] # Question을 0으로, Context를 1로 구분해 준 Segment 데이터 1번째 train_inputs[1][0] # Answer위치의 시작점과 끝점 라벨 1번째 train_labels[0][0], train_labels[1][0]<jupyter_output><empty_output><jupyter_text> ## LSTM 모델 학습 KorQuAD 태스크를 LSTM 모델을 활용하여 학습해 봅시다. 다소 복잡해 보이겠지만 Input이 2개, Output이 2개인 모델이라는 점에 주목해 주십시오. 2개의 Input은 이전 스텝에서 보았던 ```train_inputs[0], train_inputs[1]``` 이 들어갑니다. 이들은 각각 Question+Context의 데이터와 Segment입니다. 그리고 Output은 Answer의 시작점과 끝점의 위치입니다. <jupyter_code># LSTM 모델 생성 함수 작성 def build_model_lstm(n_vocab, n_seq, d_model): tokens = tf.keras.layers.Input((None,), name='tokens') segments = tf.keras.layers.Input((None,), name='segments') hidden = tf.keras.layers.Embedding(n_vocab, d_model)(tokens) + tf.keras.layers.Embedding(2, d_model)(segments) # (bs, n_seq, d_model) hidden = tf.keras.layers.LSTM(d_model, return_sequences=True)(hidden) # (bs, n_seq, d_model) hidden = tf.keras.layers.LSTM(d_model, return_sequences=True)(hidden) # (bs, n_seq, d_model) hidden = tf.keras.layers.Dense(2)(hidden) # (bs, n_seq, 2) start_logits, end_logits = tf.split(hidden, 2, axis=-1) # (bs, n_seq, 1), (bs, n_seq, 1) start_logits = tf.squeeze(start_logits, axis=-1) # (bs, n_seq) start_outputs = tf.keras.layers.Softmax(name="start")(start_logits) end_logits = tf.squeeze(end_logits, axis=-1) # (bs, n_seq) end_outputs = tf.keras.layers.Softmax(name="end")(end_logits) model = tf.keras.Model(inputs=(tokens, segments), outputs=(start_outputs, end_outputs)) return model # 모델 생성 model = build_model_lstm(n_vocab=len(vocab), n_seq=512, d_model=512) tf.keras.utils.plot_model(model, 'model.png', show_shapes=True) # 모델 컴파일 model.compile(loss=tf.keras.losses.sparse_categorical_crossentropy, optimizer=tf.keras.optimizers.Adam(learning_rate=5e-4), metrics=["accuracy"]) # 모델 학습 # 1 Epoch 에 3분 가량 소요 # 20 Epoch 진행. 5 Epoch 이상 val_start_accuracy 향상이 없으면 early stopping # early stopping early_stopping = tf.keras.callbacks.EarlyStopping(monitor='val_start_accuracy', patience=5) # save weights save_weights = tf.keras.callbacks.ModelCheckpoint(os.path.join(data_dir, "korquad_lstm.hdf5"), monitor='val_start_accuracy', verbose=1, save_best_only=True, mode='max', save_freq='epoch', save_weights_only=True) history = model.fit(train_inputs, train_labels, epochs=20, batch_size=128, validation_data=(dev_inputs, dev_labels), callbacks=[early_stopping, save_weights])<jupyter_output>Epoch 1/20 469/469 [==============================] - ETA: 0s - loss: 9.1249 - start_loss: 4.4189 - end_loss: 4.7060 - start_accuracy: 0.0685 - end_accuracy: 0.0526 Epoch 00001: val_start_accuracy improved from -inf to 0.09147, saving model to /home/ssac29/aiffel/bert_qna/data/korquad_lstm.hdf5 469/469 [==============================] - 246s 525ms/step - loss: 9.1249 - start_loss: 4.4189 - end_loss: 4.7060 - start_accuracy: 0.0685 - end_accuracy: 0.0526 - val_loss: 8.2432 - val_start_loss: 3.9167 - val_end_loss: 4.3265 - val_start_accuracy: 0.0915 - val_end_accuracy: 0.0797 Epoch 2/20 469/469 [==============================] - ETA: 0s - loss: 7.2974 - start_loss: 3.4785 - end_loss: 3.8189 - start_accuracy: 0.1254 - end_accuracy: 0.1168 Epoch 00002: val_start_accuracy improved from 0.09147 to 0.09867, saving model to /home/ssac29/aiffel/bert_qna/data/korquad_lstm.hdf5 469/469 [==============================] - 247s 528ms/step - loss: 7.2974 - start_loss: 3.4785 - end_loss: 3.8189 - star[...]<jupyter_text> ### 모델 학습 결과 시각화 <jupyter_code># training result plt.figure(figsize=(16, 4)) plt.subplot(1, 3, 1) plt.plot(history.history['loss'], 'b-', label='loss') plt.plot(history.history['val_loss'], 'r--', label='val_loss') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 3, 2) plt.plot(history.history['start_accuracy'], 'g-', label='start_accuracy') plt.plot(history.history['val_start_accuracy'], 'k--', label='val_start_accuracy') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 3, 3) plt.plot(history.history['end_accuracy'], 'b-', label='end_accuracy') plt.plot(history.history['val_end_accuracy'], 'g--', label='val_end_accuracy') plt.xlabel('Epoch') plt.legend() plt.show()<jupyter_output><empty_output><jupyter_text> ### LSTM 모델 학습 결과 LSTM 모델을 통한 학습 결과 val_loss가 낮아지지 않고, val_accuracy들도 크게 좋아지지 않는 것을 보실 수 있습니다. KorQuAD 태스크는 데이터셋만 가지고 사전 준비 없이 학습했을 때 일정 이상 좋아지지 않는다는 것을 알 수 있습니다. 모델을 다양하게 바꾸어 보아도 결과는 비슷할 것입니다. ## BERT 모델의 구조 ### Transformer 모델 Transformer 모델은 Self-Attention 으로 이루어진 __Encoder Network__ 와 __Decoder Network__ 의 구조를 갖고 있습니다. 이 Encoder-Decoder 구조로 인해 '번역기' 구현에 적합한 모델 형태입니다. ### BERT 모델 BERT 모델은 Self-Attention 으로 이루어진 Transformer __Encoder Network__ 구조만을 갖습니다. Layer 개수를 12개 이상으로 늘리고, 전체적으로 파라미터의 크기가 훨씬 큽니다. ## BERT 모델의 특징 참고. [BERT 논문정리](https://mino-park7.github.io/nlp/2018/12/12/bert-%EB%85%BC%EB%AC%B8%EC%A0%95%EB%A6%AC/?fbclid=IwAR3S-8iLWEVG6FGUVxoYdwQyA-zG0GpOUzVEsFBd0ARFg4eFXqCyGLznu7w#33-pre-training-tasks) ### Transfer Learning (전이학습) 을 활용하는 방법론 Transfer Learning, 전이학습은 대용량 unlabeld data 로 모델을 미리 학습 시킨 후, 이 학습된 모델을 불러와 특정 태스크를 갖고 있는 labeled data 로 다시한번 모델을 학습시키는 방법을 말합니다. 전이학습을 통해 token-level 태스크 부터 question-answer 태스크 까지 광범한 NLP 태스크들이 많을 성능 향상을 보였습니다. - __feature-based approach 전이학습__ OpenAI GPT 나 ELMo 등 기존의 전이학습은 두 개의 Network 를 붙여서 pre-trained lenguage representation 을 적용하였습니다. pre-trained 모델을 불러와 뒤쪽에 특정 태스크를 처리하는 network 를 더하여 새로운 모델을 생성하여 학습하였습니다. 이러한 접근 방식은 입출력 sequence 에 대해 단방향적(unidirctional) 성질로 인해 language representation 이 부족합니다. - __fine-tuning approach 전이학습__ 이와 달리 BERT 모델의 전이학습은 한 개의 Network 안에서 pre-trained 된 parameter 들을 downstream 학습을 통해 적용하였습니다. pre-trained 모델에 특정 태스크를 처리하기위한 network 를 더하지 않고, BERT 모델 자체의 (parameter) fine-tuning 을 통해 특정 태스크를 처리하기 때문에 학습 parameter 의 개수가 획기적으로 줄어들 수 있습니다 (?) ### BERT 모델의 pre-training 방법 : fine-tuning approach pre-training 은 NLP 의 여러 태스크들 중 '두 문장 사이의 관계를 이해하는 것이 중요한 경우'에 필요합니다. 기존의 OpenAI GPT 나 ELMo 모델의 경우 pre-trining 방법은 앞의 n 개 단어로 뒤의 단어를 예측하는 n-gram 모델인 일반적인 Language Model 을 사용합니다. 이러한 방식은 단방향적인 (unidirectional) 단점이 있습니다. BERT pre-training 방법은 __Masked Language Model (MLM)__ 과 __Next Sentence Prediction (NSP)__ 두 가지로 이루어집니다. - __Masked Language Model(MLM)__ 입력데이터가 \[나는\] \[MASK\] \[먹었다\] 일 때 BERT 모델이 \[MASK\] 가 \[밥을\] 임을 맞출 수 있도록 하는 언어 모델입니다. Next Token Prediction Language Model 과 대비해 '__다음 빈칸에 알맞은 말은 문제__'를 학습하는 언어 모델을 구현한 것입니다. MLM 은 Transformer Network 의 input data 에서 무작위하게 몇개의 token 을 mask 시킵니다. 그리고 이를 Transformer 구조에 넣어서 주변 단어의 context만을 보고 mask 된 단어 token 을 예측하는 모델입니다. 이는 left-to-right 를 통해 문장 전체를 predict 하는 방법과 달리, mask 된 toekn 만을 predict 하는 pre-training 태스크를 수행합니다. \[MASK\] token은 pre-training 에만 사용되고, fine-tuning시에는 사용되지 않습니다. 해당 token 을 맞추어 내는 태스크를 수행하면서, 문맥을 파악하는 능력을 길러내게 됩니다. input data 전체와 mask된 token을 한번에 Transformer encoder에 넣고 원래 token 값을 예측하므로 deep bidirectional 하다고 할 수 있습니다. - __Next Sentence Prediction__ 입력데이터가 \[나는\] \[밥을\] \[먹었다.\] \[SEP\] \[그래서\] \[지금\] \[배가\] \[부르다.\] 가 주어졌을 때, \[SEP\] 를 경계로 좌우 두 문장이 순서대로 이어지는 문장이 맞는지를 맞추는 문제입니다. BERT 모델은 이 두 문장을 입력으로 받았을 때 첫 번째 바이트에 NSP 결과값 \[CLS\](?) 을 리턴하게 됩니다. NSP 는 pre-training 시에 두 문장을 같이 넣어줘서 두 문장이 이어지는 문장인지 아닌지 맞추는 것입니다. 50:50 비율로 실제로 이어지는 두 문장과 랜덤하게 추출된 두 문장을 넣어줘서 BERT 모델이 맞추게 학습 시킵니다. input data 에 들어가는 두 문장의 token 들 사이에 \[SEP\] 토큰을 넣어 줍니다. ### BERT 모델의 입력 (input date) BERT 모델의 입력 부분을 주목해 볼 필요가 있습니다. __Input data__ $\quad$ | \[CLS\] | my | dog | is | cute | \[SEP\] | he | likes | play | ##ing | \[SEP\] --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- $\quad$__TokenEmbeddings__ $\quad$ | E$_{[CLS]}$ | E$_{my}$ | E$_{dog}$ | E$_{is}$ | E$_{cuet}$ | E$_{[ESP]}$ | E$_{he}$ | E$_{lieks}$ | E$_{play}$ | E$_{\#\#ing}$ | E$_{[SEP]}$ $\quad$__SegmentEmbeddings__ $\quad$ | E$_{A}$ | E$_{A}$ | E$_{A}$ | E$_{A}$ | E$_{A}$ | E$_{A}$ | E$_{B}$ | E$_{B}$ | E$_{B}$ | E$_{B}$ | E$_{B}$ $\quad$__PositionEmbeddings__ $\quad$ | E$_{0}$ | E$_{1}$ | E$_{2}$ | E$_{3}$ | E$_{4}$ | E$_{5}$ | E$_{6}$ | E$_{7}$ | E$_{8}$ | E$_{9}$ | E$_{10}$ 텍스트 입력이 위 그림의 \[Input\] 처럼 주어졌을 때, 실제로 모델에 입력되는 것은 __Token Embedding__, __Segment Embedding__, __Position Embedding__ 3가지가 더해진 형태입니다. ( 실제로는 그 이후 Layer Normalization 과 Dropout 이 추가로 적용됩니다 ) - __Token Embedding__ BERT는 텍스트의 tokenizer 로 Word Piece model 이라는 subword(형태소) tokenizer를 사용합니다. 문자(char) 단위로 임베딩하는 것이 기본이지만, 자주 등장하는 긴 길이의 subword(형태소)도 하나의 단위로 만들어 줍니다. 자주 등장하지 않는 단어는 다시 subword(형태소) 단위로 쪼개집니다. 이것은 자주 등장하지 않는 단어가 OOV(Out-of-vocabulary) 처리되는 것을 방지해 주는 장점도 있습니다. 최종적으로 Word Piece 모델의 각 임베딩이 입력됩니다. - __Segment Embedding__ 기존 Transformer 에 없던 독특한 임베딩입니다. 이것은 각 단어가 어느 문장에 포함되는지 그 역할을 규정하는 것입니다. QA 문제처럼 이 단어가 Question 문장에 속하는지, Context 문장에 속하는지 구분이 필요한 경우 이 임베딩은 유용하게 사용됩니다. - __Position Embedding__ 단위 토큰이 문장 내 어느 위치에 있는지를 임베딩합니다. 이 임베딩은 기존의 Transformer에서 사용되던 position embedding과 동일합니다. ## BERT 모델 구현 ### BERT 모델을 구성하는 레이어 준비 <jupyter_code># 유틸리티 함수들 def get_pad_mask(tokens, i_pad=0): """ pad mask 계산하는 함수 :param tokens: tokens (bs, n_seq) :param i_pad: id of pad :return mask: pad mask (pad: 1, other: 0) """ mask = tf.cast(tf.math.equal(tokens, i_pad), tf.float32) mask = tf.expand_dims(mask, axis=1) return mask def get_ahead_mask(tokens, i_pad=0): """ ahead mask 계산하는 함수 :param tokens: tokens (bs, n_seq) :param i_pad: id of pad :return mask: ahead and pad mask (ahead or pad: 1, other: 0) """ n_seq = tf.shape(tokens)[1] ahead_mask = 1 - tf.linalg.band_part(tf.ones((n_seq, n_seq)), -1, 0) ahead_mask = tf.expand_dims(ahead_mask, axis=0) pad_mask = get_pad_mask(tokens, i_pad) mask = tf.maximum(ahead_mask, pad_mask) return mask @tf.function(experimental_relax_shapes=True) def gelu(x): """ gelu activation 함수 :param x: 입력 값 :return: gelu activation result """ return 0.5 * x * (1 + K.tanh(x * 0.7978845608 * (1 + 0.044715 * x * x))) def kernel_initializer(stddev=0.02): """ parameter initializer 생성 :param stddev: 생성할 랜덤 변수의 표준편차 """ return tf.keras.initializers.TruncatedNormal(stddev=stddev) def bias_initializer(): """ bias initializer 생성 """ return tf.zeros_initializer class Config(dict): """ json을 config 형태로 사용하기 위한 Class :param dict: config dictionary """ __getattr__ = dict.__getitem__ __setattr__ = dict.__setitem__ @classmethod def load(cls, file): """ file에서 Config를 생성 함 :param file: filename """ with open(file, 'r') as f: config = json.loads(f.read()) return Config(config) # mode == "embedding" 일 경우 Token Embedding Layer 로 사용되는 layer 클래스입니다. class SharedEmbedding(tf.keras.layers.Layer): """ Weighed Shared Embedding Class """ def __init__(self, config, name="weight_shared_embedding"): """ 생성자 :param config: Config 객체 :param name: layer name """ super().__init__(name=name) self.n_vocab = config.n_vocab self.d_model = config.d_model def build(self, input_shape): """ shared weight 생성 :param input_shape: Tensor Shape (not used) """ with tf.name_scope("shared_embedding_weight"): self.shared_weights = self.add_weight( "weights", shape=[self.n_vocab, self.d_model], initializer=kernel_initializer() ) def call(self, inputs, mode="embedding"): """ layer 실행 :param inputs: 입력 :param mode: 실행 모드 :return: embedding or linear 실행 결과 """ # mode가 embedding일 경우 embedding lookup 실행 if mode == "embedding": return self._embedding(inputs) # mode가 linear일 경우 linear 실행 elif mode == "linear": return self._linear(inputs) # mode가 기타일 경우 오류 발생 else: raise ValueError(f"mode {mode} is not valid.") def _embedding(self, inputs): """ embedding lookup :param inputs: 입력 """ embed = tf.gather(self.shared_weights, tf.cast(inputs, tf.int32)) return embed def _linear(self, inputs): # (bs, n_seq, d_model) """ linear 실행 :param inputs: 입력 """ n_batch = tf.shape(inputs)[0] n_seq = tf.shape(inputs)[1] inputs = tf.reshape(inputs, [-1, self.d_model]) # (bs * n_seq, d_model) outputs = tf.matmul(inputs, self.shared_weights, transpose_b=True) outputs = tf.reshape(outputs, [n_batch, n_seq, self.n_vocab]) # (bs, n_seq, n_vocab) return outputs class PositionalEmbedding(tf.keras.layers.Layer): """ Positional Embedding Class """ def __init__(self, config, name="position_embedding"): """ 생성자 :param config: Config 객체 :param name: layer name """ super().__init__(name=name) self.embedding = tf.keras.layers.Embedding(config.n_seq, config.d_model, embeddings_initializer=kernel_initializer()) def call(self, inputs): """ layer 실행 :param inputs: 입력 :return embed: positional embedding lookup 결과 """ position = tf.cast(tf.math.cumsum(tf.ones_like(inputs), axis=1, exclusive=True), tf.int32) embed = self.embedding(position) return embed class ScaleDotProductAttention(tf.keras.layers.Layer): """ Scale Dot Product Attention Class """ def __init__(self, name="scale_dot_product_attention"): """ 생성자 :param name: layer name """ super().__init__(name=name) def call(self, Q, K, V, attn_mask): """ layer 실행 :param Q: Q value :param K: K value :param V: V value :param attn_mask: 실행 모드 :return attn_out: attention 실행 결과 """ attn_score = tf.matmul(Q, K, transpose_b=True) scale = tf.math.sqrt(tf.cast(tf.shape(K)[-1], tf.float32)) attn_scale = tf.math.divide(attn_score, scale) attn_scale -= 1.e9 * attn_mask attn_prob = tf.nn.softmax(attn_scale, axis=-1) attn_out = tf.matmul(attn_prob, V) return attn_out class MultiHeadAttention(tf.keras.layers.Layer): """ Multi Head Attention Class """ def __init__(self, config, name="multi_head_attention"): """ 생성자 :param config: Config 객체 :param name: layer name """ super().__init__(name=name) self.d_model = config.d_model self.n_head = config.n_head self.d_head = config.d_head # Q, K, V input dense layer self.W_Q = tf.keras.layers.Dense(config.n_head * config.d_head, kernel_initializer=kernel_initializer(), bias_initializer=bias_initializer()) self.W_K = tf.keras.layers.Dense(config.n_head * config.d_head, kernel_initializer=kernel_initializer(), bias_initializer=bias_initializer()) self.W_V = tf.keras.layers.Dense(config.n_head * config.d_head, kernel_initializer=kernel_initializer(), bias_initializer=bias_initializer()) # Scale Dot Product Attention class self.attention = ScaleDotProductAttention(name="self_attention") # output dense layer self.W_O = tf.keras.layers.Dense(config.d_model, kernel_initializer=kernel_initializer(), bias_initializer=bias_initializer()) def call(self, Q, K, V, attn_mask): """ layer 실행 :param Q: Q value :param K: K value :param V: V value :param attn_mask: 실행 모드 :return attn_out: attention 실행 결과 """ # reshape Q, K, V, attn_mask batch_size = tf.shape(Q)[0] Q_m = tf.transpose(tf.reshape(self.W_Q(Q), [batch_size, -1, self.n_head, self.d_head]), [0, 2, 1, 3]) # (bs, n_head, Q_len, d_head) K_m = tf.transpose(tf.reshape(self.W_K(K), [batch_size, -1, self.n_head, self.d_head]), [0, 2, 1, 3]) # (bs, n_head, K_len, d_head) V_m = tf.transpose(tf.reshape(self.W_V(V), [batch_size, -1, self.n_head, self.d_head]), [0, 2, 1, 3]) # (bs, n_head, K_len, d_head) attn_mask_m = tf.expand_dims(attn_mask, axis=1) # Scale Dot Product Attention with multi head Q, K, V, attn_mask attn_out = self.attention(Q_m, K_m, V_m, attn_mask_m) # (bs, n_head, Q_len, d_head) # transpose and liner attn_out_m = tf.transpose(attn_out, perm=[0, 2, 1, 3]) # (bs, Q_len, n_head, d_head) attn_out = tf.reshape(attn_out_m, [batch_size, -1, config.n_head * config.d_head]) # (bs, Q_len, d_model) attn_out = self.W_O(attn_out) # (bs, Q_len, d_model) return attn_out class PositionWiseFeedForward(tf.keras.layers.Layer): """ Position Wise Feed Forward Class """ def __init__(self, config, name="feed_forward"): """ 생성자 :param config: Config 객체 :param name: layer name """ super().__init__(name=name) self.W_1 = tf.keras.layers.Dense(config.d_ff, activation=gelu, kernel_initializer=kernel_initializer(), bias_initializer=bias_initializer()) self.W_2 = tf.keras.layers.Dense(config.d_model, kernel_initializer=kernel_initializer(), bias_initializer=bias_initializer()) def call(self, inputs): """ layer 실행 :param inputs: inputs :return ff_val: feed forward 실행 결과 """ ff_val = self.W_2(self.W_1(inputs)) return ff_val class EncoderLayer(tf.keras.layers.Layer): """ Encoder Layer Class """ def __init__(self, config, name="encoder_layer"): """ 생성자 :param config: Config 객체 :param name: layer name """ super().__init__(name=name) self.self_attention = MultiHeadAttention(config) self.norm1 = tf.keras.layers.LayerNormalization(epsilon=config.layernorm_epsilon) self.ffn = PositionWiseFeedForward(config) self.norm2 = tf.keras.layers.LayerNormalization(epsilon=config.layernorm_epsilon) self.dropout = tf.keras.layers.Dropout(config.dropout) def call(self, enc_embed, self_mask): """ layer 실행 :param enc_embed: enc_embed 또는 이전 EncoderLayer의 출력 :param self_mask: enc_tokens의 pad mask :return enc_out: EncoderLayer 실행 결과 """ self_attn_val = self.self_attention(enc_embed, enc_embed, enc_embed, self_mask) norm1_val = self.norm1(enc_embed + self.dropout(self_attn_val)) ffn_val = self.ffn(norm1_val) enc_out = self.norm2(norm1_val + self.dropout(ffn_val)) return enc_out<jupyter_output><empty_output><jupyter_text> ### BERT 모델 구현 BERT 모델의 구현에서 위에서 생성한 레이어들이 서로 결합 ! <jupyter_code>class BERT(tf.keras.layers.Layer): """ BERT Class """ def __init__(self, config, name="bert"): """ 생성자 :param config: Config 객체 :param name: layer name """ super().__init__(name=name) self.i_pad = config.i_pad self.embedding = SharedEmbedding(config) self.position = PositionalEmbedding(config) self.segment = tf.keras.layers.Embedding(2, config.d_model, embeddings_initializer=kernel_initializer()) self.norm = tf.keras.layers.LayerNormalization(epsilon=config.layernorm_epsilon) self.encoder_layers = [EncoderLayer(config, name=f"encoder_layer_{i}") for i in range(config.n_layer)] self.dropout = tf.keras.layers.Dropout(config.dropout) def call(self, enc_tokens, segments): """ layer 실행 :param enc_tokens: encoder tokens :param segments: token segments :return logits_cls: CLS 결과 logits :return logits_lm: LM 결과 logits """ enc_self_mask = get_pad_mask(enc_tokens, self.i_pad) enc_embed = self.get_embedding(enc_tokens, segments) enc_out = self.dropout(enc_embed) for encoder_layer in self.encoder_layers: enc_out = encoder_layer(enc_out, enc_self_mask) logits_cls = enc_out[:,0] logits_lm = enc_out return logits_cls, logits_lm def get_embedding(self, tokens, segments): """ token embedding, position embedding lookup :param tokens: 입력 tokens :param segments: 입력 segments :return embed: embedding 결과 """ embed = self.embedding(tokens) + self.position(tokens) + self.segment(segments) embed = self.norm(embed) return embed<jupyter_output><empty_output><jupyter_text> ## BERT 모델 이용한 학습 아래는 BERT 레이어에 Fully Connected layer 를 붙어 KorQuAD 용으로 finetune 하기 위한 모델 클래스입니다. <jupyter_code>class BERT4KorQuAD(tf.keras.Model): def __init__(self, config): super().__init__(name='BERT4KorQuAD') self.bert = BERT(config) self.dense = tf.keras.layers.Dense(2) def call(self, enc_tokens, segments): logits_cls, logits_lm = self.bert(enc_tokens, segments) hidden = self.dense(logits_lm) # (bs, n_seq, 2) start_logits, end_logits = tf.split(hidden, 2, axis=-1) # (bs, n_seq, 1), (bs, n_seq, 1) start_logits = tf.squeeze(start_logits, axis=-1) start_outputs = tf.keras.layers.Softmax(name="start")(start_logits) end_logits = tf.squeeze(end_logits, axis=-1) end_outputs = tf.keras.layers.Softmax(name="end")(end_logits) return start_outputs, end_outputs config = Config({"d_model": 256, "n_head": 4, "d_head": 64, "dropout": 0.1, "d_ff": 1024, "layernorm_epsilon": 0.001, "n_layer": 3, "n_seq": 384, "n_vocab": 0, "i_pad": 0}) config.n_vocab = len(vocab) config.i_pad = vocab.pad_id() config<jupyter_output><empty_output><jupyter_text> 모델의 크기가 다르고, 사용할 수 있는 배치 사이즈가 달라지므로, 배치 구성만 다시 진행 메모리 한계 때문에 배치 사이즈는 32 정도가 적당 <jupyter_code>bert_batch_size = 32 train_dataset = tf.data.Dataset.from_tensor_slices((train_inputs, train_labels)).shuffle(10000).batch(bert_batch_size) dev_dataset = tf.data.Dataset.from_tensor_slices((dev_inputs, dev_labels)).batch(bert_batch_size) model = BERT4KorQuAD(config)<jupyter_output><empty_output><jupyter_text> ### pretraining 없이 학습 BERT 는 pretrained 모델을 활용하는 데 의의가 있습니다. pretraining을 수행에는 워크스테이션급 하드웨어를 동원해서 1달 가까이 학습을 시켜야 성능이 나오는 거대한 모델입니다. 따라서 이번에는 BERT 모델만 구성한 후 전혀 pretraining 없이 학습을 진행해 보겠습니다. <jupyter_code>def train_epoch(model, dataset, loss_fn, acc_fn, optimizer): metric_start_loss = tf.keras.metrics.Mean(name='start_loss') metric_end_loss = tf.keras.metrics.Mean(name='end_loss') metric_start_acc = tf.keras.metrics.Mean(name='start_acc') metric_end_acc = tf.keras.metrics.Mean(name='end_acc') p_bar = tqdm(dataset) for batch, ((enc_tokens, segments), (start_labels, end_labels)) in enumerate(p_bar): with tf.GradientTape() as tape: start_outputs, end_outputs = model(enc_tokens, segments) start_loss = loss_fn(start_labels, start_outputs) end_loss = loss_fn(end_labels, end_outputs) loss = start_loss + end_loss start_acc = acc_fn(start_labels, start_outputs) end_acc = acc_fn(end_labels, end_outputs) gradients = tape.gradient(loss, model.trainable_variables) optimizer.apply_gradients(zip(gradients, model.trainable_variables)) metric_start_loss(start_loss) metric_end_loss(end_loss) metric_start_acc(start_acc) metric_end_acc(end_acc) if batch % 10 == 9: p_bar.set_description(f'loss: {metric_start_loss.result():0.4f}, {metric_end_loss.result():0.4f}, acc: {metric_start_acc.result():0.4f}, {metric_end_acc.result():0.4f}') p_bar.close() return metric_start_loss.result(), metric_end_loss.result(), metric_start_acc.result(), metric_end_acc.result() def eval_epoch(model, dataset, loss_fn, acc_fn): metric_start_loss = tf.keras.metrics.Mean(name='start_loss') metric_end_loss = tf.keras.metrics.Mean(name='end_loss') metric_start_acc = tf.keras.metrics.Mean(name='start_acc') metric_end_acc = tf.keras.metrics.Mean(name='end_acc') for batch, ((enc_tokens, segments), (start_labels, end_labels)) in enumerate(dataset): start_outputs, end_outputs = model(enc_tokens, segments) start_loss = loss_fn(start_labels, start_outputs) end_loss = loss_fn(end_labels, end_outputs) start_acc = acc_fn(start_labels, start_outputs) end_acc = acc_fn(end_labels, end_outputs) metric_start_loss(start_loss) metric_end_loss(end_loss) metric_start_acc(start_acc) metric_end_acc(end_acc) return metric_start_loss.result(), metric_end_loss.result(), metric_start_acc.result(), metric_end_acc.result() # 모델 학습 # 1 Epoch 에 5분 가량 소요 # 20 Epoch 진행. 5 Epoch 이상 val_start_accuracy 향상이 없으면 early stopping train_history = [] eval_history = [] loss_fn = tf.keras.losses.sparse_categorical_crossentropy acc_fn = tf.keras.metrics.sparse_categorical_accuracy optimizer = tf.keras.optimizers.Adam(learning_rate=5e-4) best_acc = .0 patience = 0 for epoch in range(20): train_start_loss, train_end_loss, train_start_acc, train_end_acc = train_epoch(model, train_dataset, loss_fn, acc_fn, optimizer) start_loss, end_loss, start_acc, end_acc = eval_epoch(model, dev_dataset, loss_fn, acc_fn) train_result = [epoch, train_start_loss, train_end_loss, train_start_acc, train_end_acc] eval_result = [epoch, start_loss, end_loss, start_acc, end_acc] print(f'train {epoch} >> loss: {train_start_loss:0.4f}, {train_end_loss:0.4f}, acc: {train_start_acc:0.4f}, {train_end_acc:0.4f}') print(f'eval {epoch} >> loss: {start_loss:0.4f}, {end_loss:0.4f}, acc: {start_acc:0.4f}, {end_acc:0.4f}') acc = start_acc + end_acc if best_acc < acc: patience = 0 best_acc = acc model.save_weights(os.path.join(data_dir, "korquad_bert_pretrained.hdf5")) print(f'save best model') else: patience += 1 if 5 <= patience: print(f'early stopping') break<jupyter_output><empty_output><jupyter_text> ### 모델 학습 결과 시각화 <jupyter_code># 학습 과정에서의 모델 매트릭 log 확인 # vanilla BERT 모델 학습결과 train loss 와 acc, validation loss 와 acc 를 저장 vanilla_train_start_loss = [4.0624, 3.4155, 3.0915, 2.8805, 2.7345, 2.4980, 2.2642, 1.9725, 1.6739, 1.3955] vanilla_train_end_loss = [4.5772, 3.9612, 3.5802, 3.2991, 3.0948, 2.7852, 2.4885, 2.1004, 1.7605, 1.4605] vanilla_train_start_acc = [0.0812, 0.1397, 0.1945, 0.2313, 0.2593, 0.3089, 0.3651, 0.4346, 0.5152, 0.5880] vanilla_train_end_acc = [0.0610, 0.1158, 0.1663, 0.2105, 0.2469, 0.3039, 0.3682, 0.4570, 0.5328, 0.6045] vanilla_eval_start_loss = [3.9064, 3.6582, 3.6936, 3.6640, 3.9409, 4.0885, 4.3037, 4.8327, 5.1923, 5.3824] vanilla_eval_end_loss = [4.4222, 4.2336, 4.2645, 4.2235, 4.4810, 4.5909, 4.7771, 5.3338, 6.1183, 6.0056] vanilla_eval_start_acc = [0.0967, 0.1410, 0.1584, 0.1619, 0.1573, 0.1592, 0.1533, 0.1573, 0.1496, 0.1466] vanilla_eval_end_acc = [0.0720, 0.1178, 0.1390, 0.1440, 0.1520, 0.1480, 0.1448, 0.1519, 0.1443, 0.1422] # training result # 위 학습 log 의 loss, acc 데이터 리스트 만들어서 시각화 합시다 # 다음번 모델 학습 때는 리스트 따로 만들어서 관리 합시다. 저장파일을 만들던 ... plt.figure(figsize=(16, 4)) plt.subplot(1, 4, 1) plt.plot(vanilla_train_start_loss, 'b--', label='vanilla_train_start_loss') plt.plot(vanilla_eval_start_loss, 'r-', label='vanilla_eval_start_loss') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 4, 2) plt.plot(vanilla_train_end_loss, 'b--', label='vanilla_train_end_loss') plt.plot(vanilla_eval_end_loss, 'r-', label='vanilla_eval_end_loss') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 4, 3) plt.plot(vanilla_train_start_acc, 'b--', label='vanilla_train_start_acc') plt.plot(vanilla_eval_start_acc, 'r-', label='vanilla_eval_start_acc') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 4, 4) plt.plot(vanilla_train_end_acc, 'b--', label='vanilla_train_end_acc') plt.plot(vanilla_eval_end_acc, 'r-', label='vanilla_eval_end_acc') plt.xlabel('Epoch') plt.legend() plt.show() dev_json = os.path.join(data_dir, "korquad_dev.json") with open(dev_json) as f: cnt = 0 for i, line in enumerate(f): data = json.loads(line) question = vocab.decode_pieces(data['question']) context = vocab.decode_pieces(data['context']) answer = data['answer'] answer_predict = do_predict(model, question, context) if answer in answer_predict: print(i) print("질문 : ", question) print("지문 : ", context) print("정답 : ", answer) print("예측 : ", answer_predict, "\n") else: print('예측 실패!', "\n") if 100 < i: break<jupyter_output>예측 실패! 예측 실패! 2 질문 : 임종석이 여의도 농민 폭력 시위를 주도한 혐의로 지명수배된 연도는? 지문 : 1989년 2월 15일 여의도 농민 폭력 시위를 주도한 혐의(폭력행위등처벌에관한법률위반)으로 지명수배되었다. 1989년 3월 12일 서울지방검찰청 공안부는 임종석의 사전구속영장을 발부받았다. 같은 해 6월 30일 평양축전에 임수경을 대표로 파견하여 국가보안법위반 혐의가 추가되었다. 경찰은 12월 18일~20일 사이 서울 경희대학교에서 임종석이 성명 발표를 추진하고 있다는 첩보를 입수했고, 12월 18일 오전 7시 40분 경 가스총과 전자봉으로 무장한 특공조 및 대공과 직원 12명 등 22명의 사복 경찰을 승용차 8대에 나누어 경희대학교에 투입했다. 1989년 12월 18일 오전 8시 15분 경 서울청량리경찰서는 호위 학생 5명과 함께 경희대학교 학생회관 건물 계단을 내려오는 임종석을 발견, 검거해 구속을 집행했다. 임종석은 청량리경찰서에서 약 1시간 동안 조사를 받은 뒤 오전 9시 50분 경 서울 장안동의 서울지방경찰청 공안분실로 인계되었다. 정답 : 1989년 예측 : 1989년 예측 실패! 예측 실패! 예측 실패! 예측 실패! 7 질문 : 정부의 헌법개정안 준비 과정에 대해서 청와대 비서실이 아니라 국무회의 중심으로 이뤄졌어야 했다고 지적한 원로 헌법학자는? 지문 : "내각과 장관들이 소외되고 대통령비서실의 권한이 너무 크다", "행보가 비서 본연의 역할을 벗어난다"는 의견이 제기되었다. 대표적인 예가 10차 개헌안 발표이다. 원로 헌법학자인 허영 경희대 석좌교수는 정부의 헌법개정안 준비 과정에 대해 "청와대 비서실이 아닌 국무회의 중심으로 이뤄졌어야 했다"고 지적했다. '국무회의의 심의를 거쳐야 한다'(제89조)는 헌법 규정에 충실하지 않았다는 것이다. 그러면서 "법무부 장관을 제쳐놓고 민정수석이 개정안을 설명하는 게 이해가 안 된다"고 지적했다. 민정수석은 국회의원에 대해 책임지는 법무부 장관도 아니고, [...]<jupyter_text> ### pretraining 없이 BERT 모델 학습 결과 pretraining 없이 BERT 모델을 통한 학습 결과 Epoch 가 진행될 수록 train accuracy 는 0.45 까지 좋아지는 반면 validation accuracy 는 0.13 대에 머물며 나아지지 않는 것을 보실 수 있습니다. 이 모델에는 수많은 코퍼스를 통해 정교하게 얻어진 Word Embedding 이 반영되지 않았기 때문입니다. ## Pretrained BERT 모델 사용 ### pretrained model 불러오기 <jupyter_code>checkpoint_file = os.path.join(model_dir, 'bert_pretrain_32000.hdf5') model = BERT4KorQuAD(config) if os.path.exists(checkpoint_file): # pretrained model 을 로드하기 위해 먼저 모델이 생성되어 있어야 한다. enc_tokens = np.random.randint(0, len(vocab), (4, 10)) segments = np.random.randint(0, 2, (4, 10)) model(enc_tokens, segments) # checkpoint 파일로부터 필요한 layer를 불러온다. model.load_weights(os.path.join(model_dir, "bert_pretrain_32000.hdf5"), by_name=True) model.summary() else: print('NO Pretrained Model')<jupyter_output>Model: "BERT4KorQuAD" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= bert (BERT) multiple 10662400 _________________________________________________________________ dense_76 (Dense) multiple 514 ================================================================= Total params: 10,662,914 Trainable params: 10,662,914 Non-trainable params: 0 _________________________________________________________________ <jupyter_text> ### pretrained model finetune 학습을 진행하는 코드는 이전 스텝과 동일합니다. 단지 학습해야 할 모델이 랜덤 초기화된 것이 아니라 pretrained model 을 로드한 것일 뿐입니다. ### fintune 학습이 완료된 모델로 Test (inference 수행) finetune 학습이 완료된 model을 활용하여 실제 퀴즈 풀이 결과를 확인해 봅니다. <jupyter_code>def do_predict(model, question, context): """ 입력에 대한 답변 생성하는 함수 :param model: model :param question: 입력 문자열 :param context: 입력 문자열 """ q_tokens = vocab.encode_as_pieces(question)[:args.max_query_length] c_tokens = vocab.encode_as_pieces(context)[:args.max_seq_length - len(q_tokens) - 3] tokens = ['[CLS]'] + q_tokens + ['[SEP]'] + c_tokens + ['[SEP]'] token_ids = [vocab.piece_to_id(token) for token in tokens] segments = [0] * (len(q_tokens) + 2) + [1] * (len(c_tokens) + 1) y_start, y_end = model(np.array([token_ids]), np.array([segments])) # print(y_start, y_end) y_start_idx = K.argmax(y_start, axis=-1)[0].numpy() y_end_idx = K.argmax(y_end, axis=-1)[0].numpy() answer_tokens = tokens[y_start_idx:y_end_idx + 1] return vocab.decode_pieces(answer_tokens)<jupyter_output><empty_output><jupyter_text> ### 학습된 모델로 예측 테스트 <jupyter_code>dev_json = os.path.join(data_dir, "korquad_dev.json") with open(dev_json) as f: for i, line in enumerate(f): data = json.loads(line) question = vocab.decode_pieces(data['question']) context = vocab.decode_pieces(data['context']) answer = data['answer'] answer_predict = do_predict(model, question, context) if answer in answer_predict: print(i) print("질문 : ", question) print("지문 : ", context) print("정답 : ", answer) print("예측 : ", answer_predict, "\n") if 100 < i: break # 모델 학습 # 1 Epoch 에 5분 가량 소요 # 20 Epoch 진행. 5 Epoch 이상 val_start_accuracy 향상이 없으면 early stopping loss_fn = tf.keras.losses.sparse_categorical_crossentropy acc_fn = tf.keras.metrics.sparse_categorical_accuracy optimizer = tf.keras.optimizers.Adam(learning_rate=5e-4) best_acc = .0 patience = 0 for epoch in range(20): train_start_loss, train_end_loss, train_start_acc, train_end_acc = train_epoch(model, train_dataset, loss_fn, acc_fn, optimizer) start_loss, end_loss, start_acc, end_acc = eval_epoch(model, dev_dataset, loss_fn, acc_fn) train_result = [epoch, train_start_loss, train_end_loss, train_start_acc, train_end_acc] eval_result = [epoch, start_loss, end_loss, start_acc, end_acc] print(f'train {epoch} >> loss: {train_start_loss:0.4f}, {train_end_loss:0.4f}, acc: {train_start_acc:0.4f}, {train_end_acc:0.4f}') print(f'eval {epoch} >> loss: {start_loss:0.4f}, {end_loss:0.4f}, acc: {start_acc:0.4f}, {end_acc:0.4f}') acc = start_acc + end_acc if best_acc < acc: patience = 0 best_acc = acc model.save_weights(os.path.join(data_dir, "korquad_bert_none_pretrain.hdf5")) print(f'save best model') else: patience += 1 if 5 <= patience: print(f'early stopping') break<jupyter_output><empty_output><jupyter_text> ### 모델 학습 결과 시각화 <jupyter_code># 학습 과정에서의 모델 매트릭 log 확인 # Tensor.numpy() 메서드를 사용하여 tf.Tensor 에서 값을 가져올 수 있습니다. (.array() 나 .data() 는?) pretrained_train_start_loss = [3.9208, 3.2375, 2.9960, 2.8156, 2.6055, 2.3674, 2.0659, 1.7587] pretrained_train_end_loss = [4.4334, 3.7766, 3.4050, 3.0423, 2.6455, 2.2312, 1.8056, 1.4262] pretrained_train_start_acc = [0.1028, 0.1709, 0.2071, 0.2412, 0.2829, 0.3392, 0.4116, 0.4946] pretrained_train_end_acc = [0.0839, 0.1478, 0.1990, 0.2520, 0.3219, 0.4041, 0.4988, 0.5913] pretrained_eval_start_loss = [3.6453, 3.5813, 3.6662, 3.7338, 3.8957, 4.1137, 4.5747, 4.6807] pretrained_eval_end_loss = [4.2447, 4.1463, 4.2593, 4.3728, 4.6429, 5.1066, 5.9167, 5.7736] pretrained_eval_start_acc = [0.1232, 0.1536, 0.1589, 0.1656, 0.1605, 0.1499, 0.1361, 0.1390] pretrained_eval_end_acc = [0.1066, 0.1327, 0.1434, 0.1331, 0.1338, 0.1320, 0.1229, 0.1241] # training result # 위 학습 log 의 loss, acc 데이터 리스트 만들어서 시각화 합시다 # 다음번 모델 학습 때는 리스트 따로 만들어서 관리 합시다. 저장파일을 만들던 ... plt.figure(figsize=(16, 4)) plt.subplot(1, 4, 1) plt.plot(pretrained_train_start_loss, 'b--', label='pretrained_train_start_loss') plt.plot(pretrained_eval_start_loss, 'r-', label='pretrained_eval_start_loss') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 4, 2) plt.plot(pretrained_train_end_loss, 'b--', label='pretrained_train_end_loss') plt.plot(pretrained_eval_end_loss, 'r-', label='pretrained_eval_end_loss') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 4, 3) plt.plot(pretrained_train_start_acc, 'b--', label='pretrained_train_start_acc') plt.plot(pretrained_eval_start_acc, 'r-', label='pretrained_eval_start_acc') plt.xlabel('Epoch') plt.legend() plt.subplot(1, 4, 4) plt.plot(pretrained_train_end_acc, 'b--', label='pretrained_train_end_acc') plt.plot(pretrained_eval_end_acc, 'r-', label='pretrained_eval_end_acc') plt.xlabel('Epoch') plt.legend() plt.show()<jupyter_output><empty_output><jupyter_text> ### 학습된 모델로 예측 테스트 <jupyter_code>dev_json = os.path.join(data_dir, "korquad_dev.json") with open(dev_json) as f: fail_cnt = 0 for i, line in enumerate(f): data = json.loads(line) question = vocab.decode_pieces(data['question']) context = vocab.decode_pieces(data['context']) answer = data['answer'] answer_predict = do_predict(model, question, context) if answer in answer_predict: print(i) print("질문 : ", question) print("지문 : ", context) print("정답 : ", answer) print("예측 : ", answer_predict, "\n") else: fail_cnt += 1 print('예측 실패!', "\n") if 99 <= i: pretrained_sample_test_acc = (100 - fail_cnt) / 100 break # 학습된 모델로 100 개의 샘플 테스트 예측한 것의 accuracy print(pretrained_sample_test_acc)<jupyter_output>0.18
no_license
/bert_qna.ipynb
doutoury/bert_qnd
34
<jupyter_start><jupyter_text># Preprocess Docs<jupyter_code>from preprocesing import process_data_files process_data_files.prep_anorexia_data(man_dir+"original_chunks/", man_dir+"train_golden_truth.txt", dest_dir+"prep_chunks/") man_dir="D:/corpus/DepresionEriskCollections/2017/test/" dest_dir="D:/corpus/DepresionEriskCollections/2017V2/test/" process_data_files.prep_anorexia_data(man_dir+"original_chunks/", man_dir+"test_golden_truth.txt", dest_dir+"prep_chunks/") man_dir="D:/corpus/DepresionEriskCollections/2018/" dest_dir="D:/corpus/DepresionEriskCollections/2018V2/" process_data_files.prep_anorexia_data(man_dir+"original_prep_chunks/", man_dir+"test_golden_truth.txt", dest_dir+"prep_chunks/")<jupyter_output><empty_output><jupyter_text># Read training docs<jupyter_code>from preprocesing.load_datasets import Dataset data= Dataset(key="erisk18_dev", chunking=False, remove_end=True)<jupyter_output><empty_output><jupyter_text>## Positive docs<jupyter_code>pos_docs=data.get_dataset(folder_name="prep_chunks", truth_name="train_golden_truth.txt", partition="training", chunks=False) len(pos_docs[0]) from classifier import explore_data import pandas as pd df = pd.DataFrame(pos_docs[0],index=[i for i in range (len(pos_docs[0]))], columns=["doc"]) df["truth"]=pos_docs[1] pos_docs_1=list (df[df["truth"]==1]["doc"]) len(pos_docs_1) explore_data.plot_sample_length_distribution(pos_docs_1) explore_data.plot_frequency_distribution_of_ngrams(pos_docs_1, (1,1), 20)<jupyter_output><empty_output><jupyter_text>## Negative docs<jupyter_code>pos_docs_0=list (df[df["truth"]==0]["doc"]) explore_data.plot_sample_length_distribution(pos_docs_0) explore_data.plot_frequency_distribution_of_ngrams(pos_docs_0, (1,1), 20)<jupyter_output><empty_output><jupyter_text>## Both clases<jupyter_code>explore_data.plot_class_distribution(pos_docs[1]) explore_data.plot_sample_length_distribution(pos_docs[0]) explore_data.plot_frequency_distribution_of_ngrams(pos_docs[0], (1,1), 20) explore_data.plot_frequency_distribution_of_ngrams(pos_docs[0], (2,2), 20) explore_data.plot_frequency_distribution_of_ngrams(pos_docs[0], (3,3), 20)<jupyter_output><empty_output><jupyter_text># Clasificacion y ganancia de informacion<jupyter_code>from classifier.SVM_Text import SVM_text from classifier.FeactureExtraction import feature_extraction svm_obj=SVM_text(pos_docs[0], pos_docs[1], weighted=True)<jupyter_output><empty_output><jupyter_text>## Cargar conjunto de test<jupyter_code>svm_obj.extract_features(pos_docs[0], idf=True, stop_words="english", norm="l2") svm_obj.train_and_test(pos_docs[1], p_label=1) svm_obj.ft.get_chi_2(pos_docs[1], 100)<jupyter_output><empty_output><jupyter_text>## Validation<jupyter_code>val=data.get_dataset(folder_name="prep_chunks", truth_name="test_golden_truth.txt", partition="test", chunks=False) len(val[1]) svm_obj.extract_features(val[0], idf=True, stop_words="english", norm="l2") svm_obj.train_and_test(val[1], p_label=1)<jupyter_output>Confusion matrix: [[328 21] [ 23 29]] Scores: [136467, 0.8902743142144638, 0.58, 0.5576923076923077, 0.5686274509803922] <jupyter_text># Test real<jupyter_code>data2= Dataset(key="erisk18_test", chunking=False, remove_end=True) test=data2.get_dataset(folder_name="prep_chunks", truth_name="test_golden_truth.txt", partition="test", chunks=False) svm_obj.extract_features(test[0], idf=True, stop_words="english", norm="l2") svm_obj.train_and_test(test[1], p_label=1)<jupyter_output>Confusion matrix: [[707 34] [ 37 42]] Scores: [136467, 0.9134146341463415, 0.5526315789473685, 0.5316455696202531, 0.5419354838709678]
no_license
/notebooks/preprocesamiento_depresion.ipynb
v1ktop/data_augmentation_for_author_profiling
9
<jupyter_start><jupyter_text>Let's get coordinates! <jupyter_code>!wget -q -O 'Geospatial_Coordinates.csv' http://cocl.us/Geospatial_data print('Data downloaded!') df_geo_coor = pd.read_csv("./Geospatial_Coordinates.csv") df_geo_coor.head() df_toronto = pd.merge(df, df_geo_coor, how='left', left_on = 'PostalCode', right_on = 'Postal Code') df_toronto.drop("Postal Code", axis=1, inplace=True) df_toronto.head() address = "Toronto, ON" geolocator = Nominatim(user_agent="toronto_explorer") location = geolocator.geocode(address) latitude = location.latitude longitude = location.longitude print('The geograpical coordinate of Toronto city are {}, {}.'.format(latitude, longitude))<jupyter_output>The geograpical coordinate of Toronto city are 43.6534817, -79.3839347. <jupyter_text>Let's create the DataFrame!<jupyter_code>df_toronto_denc = df_toronto[df_toronto['Borough'].str.contains("Toronto")].reset_index(drop=True) df_toronto_denc.head()<jupyter_output><empty_output>
no_license
/Clustering Toronto.ipynb
savana2019x/Coursera_Capstone
2
<jupyter_start><jupyter_text>### Numpy learning from various tutorialsFirst tutorial: http://cs231n.github.io/python-numpy-tutorial/#numpy1. Array Creation<jupyter_code>import numpy as np a = np.array([1, 2, 3]) print(type(a)) print(a.shape) b = np.zeros((3,4,5)) print(b) c = np.ones((3,4)) print(c) d = np.full((2,2), 3) print(d) e = np.eye(4) print(e) f= np.random.rand(3) print(f)<jupyter_output>[0.21019164 0.90557905 0.71941015] <jupyter_text>2. Array Indexing<jupyter_code>np.random.seed(19) arr = np.random.rand(3,3) print(arr) print(arr[:2]) print(arr[:2,:1])<jupyter_output>[[0.0975336 ] [0.13813169]] <jupyter_text>Skipping integer array indexing from the tutorial. The basic idea from indexing is: 1. we can get single elements 2. we can get sub-arrays - through slicing - through integer arraysSkipping arithmetic from the tutuorial Skipping broadcasting too for now - will do it later in this tutuorial<jupyter_code>i = np.random.random((3,1)) print(i.shape) print(i) z = np.array([2,3]).reshape(2,1) z x = np.array([5,6]).reshape(2,1) x np.dot(x.T,z).shape<jupyter_output><empty_output>
no_license
/6.86x/tutorials/numpy.ipynb
ssinghaldev/data_science
3
<jupyter_start><jupyter_text># [【SOTA】マイナビ × SIGNATE Student Cup 2019: 賃貸物件の家賃予測](https://signate.jp/competitions/264)## 1. データ読み込み<jupyter_code>import pandas as pd import numpy as np import pathlib import os # 学習データ、テストデータの読み込み train_path = pathlib.Path("./DATA/train.csv") test_path = pathlib.Path("./DATA/test.csv") train_data = pd.read_csv(train_path) test_data = pd.read_csv(test_path) test_data.head()<jupyter_output><empty_output><jupyter_text>## 2. 前処理### 2.1. データ選択<jupyter_code>train_data_1 = train_data[["id", "アクセス", "所在地", "賃料", "間取り", "築年数", "面積", "所在階", "建物構造"]] test_data_1 = test_data[["id", "アクセス", "所在地", "間取り", "築年数", "面積", "所在階", "建物構造"]]<jupyter_output><empty_output><jupyter_text>### 2.2. 間取りの数値化<jupyter_code># 間取りにlabel encodingを適用 from sklearn.preprocessing import LabelEncoder le = LabelEncoder() le.fit(pd.concat([train_data_1["間取り"],test_data_1["間取り"]])) train_data_1["間取りID"] = le.transform(train_data_1["間取り"]) test_data_1["間取りID"] = le.transform(test_data_1["間取り"])<jupyter_output>C:\ProgramData\Anaconda3\lib\site-packages\ipykernel_launcher.py:7: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame. Try using .loc[row_indexer,col_indexer] = value instead See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy import sys C:\ProgramData\Anaconda3\lib\site-packages\ipykernel_launcher.py:8: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame. Try using .loc[row_indexer,col_indexer] = value instead See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy <jupyter_text>### 2.3. 面積の数値化<jupyter_code># 面積のカラムタイトルを面積[m2]に置換 train_data_1 = train_data_1.rename(columns={"面積": "面積m2"}) test_data_1 = test_data_1.rename(columns={"面積": "面積m2"}) # カラムタイトル面積[m2]の要素からm2を削除 train_data_1['面積m2'] = train_data_1['面積m2'].str.replace('m2', '').astype(float) test_data_1['面積m2'] = test_data_1['面積m2'].str.replace('m2', '').astype(float)<jupyter_output><empty_output><jupyter_text>### 2.4. 所在階の数値化<jupyter_code># 所在階の中身を"/"で2つの列に分割 train_data_1 = pd.concat([train_data_1, train_data_1['所在階'].str.split('/', expand=True)], axis=1) test_data_1 = pd.concat([test_data_1, test_data_1['所在階'].str.split('/', expand=True)], axis=1) # 分割した所在階のカラム名変更 0:所在、1:階層 train_data_1 = train_data_1.rename(columns={0:"所在", 1:"階層"}) test_data_1 = test_data_1.rename(columns={0:"所在", 1:"階層"}) # 所在と階層の要素を整形 train_data_1['所在'] = train_data_1['所在'].str.replace('階', '') train_data_1['所在'] = train_data_1['所在'].str.replace('建', '') train_data_1['所在'] = train_data_1['所在'].str.replace('地下', '-') train_data_1['所在'] = train_data_1['所在'].str.replace('\(.*\)', '', regex=True) train_data_1['階層'] = train_data_1['階層'].str.replace('階建', '') train_data_1['階層'] = train_data_1['階層'].str.replace('\(.*\)', '', regex=True) train_data_1['所在'] = train_data_1['所在'].replace('', np.nan) test_data_1['所在'] = test_data_1['所在'].str.replace('階', '') test_data_1['所在'] = test_data_1['所在'].str.replace('建', '') test_data_1['所在'] = test_data_1['所在'].str.replace('地下', '-') test_data_1['所在'] = test_data_1['所在'].str.replace('\(.*\)', '', regex=True) test_data_1['階層'] = test_data_1['階層'].str.replace('階建', '') test_data_1['階層'] = test_data_1['階層'].str.replace('\(.*\)', '', regex=True) test_data_1['所在'] = test_data_1['所在'].replace('', np.nan) # # "階層"がNoneの箇所を"所在"の値で埋める # train_data_1['階層'].fillna(train_data_1['所在'], inplace=True) # test_data_1['階層'].fillna(test_data_1['所在'], inplace=True) # 所在と階層の要素をfloat型に変換する train_data_1['所在'] = train_data_1['所在'].astype(float) train_data_1['階層'] = train_data_1['階層'].astype(float) test_data_1['所在'] = test_data_1['所在'].astype(float) test_data_1['階層'] = test_data_1['階層'].astype(float) # 所在階のカラムを削除 train_data_1 = train_data_1.drop('所在階', axis=1) test_data_1 = test_data_1.drop('所在階', axis=1) # # 所在も階層も空欄のデータは間取りと面積の近いデータで埋める # print(test_data_1[(test_data_1["間取りID"] == 21) & (test_data_1["面積m2"] > 90) & (test_data_1["面積m2"] < 95)].mean()) # test_data_1.loc[test_data_1["id"]==40675, "所在"] = float(6) # test_data_1.loc[test_data_1["id"]==40675, "階層"] = float(9)<jupyter_output><empty_output><jupyter_text>### 2.5. 築年数の数値化<jupyter_code># 新築の場合は全て0にする train_data_1.loc[train_data_1["築年数"]=="新築", "築年数"] = float(0) test_data_1.loc[test_data_1["築年数"]=="新築", "築年数"] = float(0) # 築年数を数値に変換する train_data_1 = pd.concat([train_data_1, train_data_1['築年数'].str.split('年', expand=True)], axis=1) test_data_1 = pd.concat([test_data_1, test_data_1['築年数'].str.split('年', expand=True)], axis=1) # ヶ月を消す train_data_1[1] = train_data_1[1].str.replace('ヶ月', '') test_data_1[1] = test_data_1[1].str.replace('ヶ月', '') # 築年数をfloat変換 train_data_1["築年数"] = train_data_1[0].astype(float) + (train_data_1[1].astype(float) / float(10)) test_data_1["築年数"] = test_data_1[0].astype(float) + (test_data_1[1].astype(float) / float(10)) test_data_1.head() print(test_data_1[test_data_1["築年数"].isnull()]) # 0と1の列を消す train_data_1 = train_data_1.drop(0, axis=1) train_data_1 = train_data_1.drop(1, axis=1) test_data_1 = test_data_1.drop(0, axis=1) test_data_1 = test_data_1.drop(1, axis=1)<jupyter_output><empty_output><jupyter_text>### 2.6. 緯度経度情報の追加#### [このサイト](https://ktgis.net/gcode/geocoding.html)を利用する<jupyter_code># IDと住所をファイル出力 train_coordinate = train_data[["id", "所在地"]] test_coordinate = test_data[["id", "所在地"]] train_coordinate.to_csv("train_coordinate.csv", header=False, index=False) test_coordinate.to_csv("test_coordinate.csv", header=False, index=False) # 緯度、経度情報の読み込み(id, 所在地, 経度, 緯度) train_coordinate_addvalue = pd.read_excel("train_coordinate_addvalue.xlsx") test_coordinate_addvalue = pd.read_excel("test_coordinate_addvalue.xlsx") train_data_1 = pd.merge(train_data_1, train_coordinate_addvalue[["id", "経度", "緯度"]], on='id') test_data_1 = pd.merge(test_data_1, test_coordinate_addvalue[["id", "経度", "緯度"]], on='id')<jupyter_output><empty_output><jupyter_text>### 2.7. 部屋数の追加<jupyter_code># 部屋数のマージ number_of_rooms = pd.read_excel("number_of_rooms.xlsx") train_data_1 = pd.merge(train_data_1, number_of_rooms[["間取り", "部屋数"]], on='間取り') test_data_1 = pd.merge(test_data_1, number_of_rooms[["間取り", "部屋数"]], on='間取り') # インデックスの振り直し train_data_1 = train_data_1.sort_values("id") train_data_1 = train_data_1.reset_index(drop=True) test_data_1 = test_data_1.sort_values("id") test_data_1 = test_data_1.reset_index(drop=True)<jupyter_output><empty_output><jupyter_text>### 2.8. 1部屋当たりの面積追加<jupyter_code># 面積と部屋数から1部屋当たりの面積を算出 train_data_1["1部屋当たり面積m2"] = (train_data_1["面積m2"] / train_data_1["部屋数"]).astype(float) test_data_1["1部屋当たり面積m2"] = (test_data_1["面積m2"] / test_data_1["部屋数"]).astype(float) # 間取りと間取りIDは削除する train_data_1 = train_data_1.drop('間取り', axis=1) test_data_1 = test_data_1.drop('間取り', axis=1) # train_data_1 = train_data_1.drop('間取りID', axis=1) # test_data_1 = test_data_1.drop('間取りID', axis=1)<jupyter_output><empty_output><jupyter_text>#### 2.9. 建物の高さ率を追加(高さ率=所在/階層)<jupyter_code># 高さ率の算出 train_data_1["高さ率"] = (train_data_1["所在"] / train_data_1["階層"]).astype(float) test_data_1["高さ率"] = (test_data_1["所在"] / test_data_1["階層"]).astype(float)<jupyter_output><empty_output><jupyter_text>### 2.10. 建物構造の数値化<jupyter_code>train_data_1["建物構造"].value_counts() # 建物構造にlabel encodingを適用 from sklearn.preprocessing import LabelEncoder le = LabelEncoder() le.fit(pd.concat([train_data_1["建物構造"],test_data_1["建物構造"]])) train_data_1["建物構造ID"] = le.transform(train_data_1["建物構造"]) test_data_1["建物構造ID"] = le.transform(test_data_1["建物構造"]) # 不要になった建物構造カラムを削除する train_data_1 = train_data_1.drop('建物構造', axis=1) test_data_1 = test_data_1.drop('建物構造', axis=1)<jupyter_output><empty_output><jupyter_text>### 2.11. 新規カラムに「賃料/面積」を追加<jupyter_code>train_data_1["賃料/面積"] = train_data_1["賃料"] / train_data_1["面積m2"]<jupyter_output><empty_output><jupyter_text>### 2.12. 東京23区の地域ごとに、住宅地の平均地価情報を追加<jupyter_code># 区の情報の確認(23区だけだった) # sample = train_data["所在地"].str.extract("東京都(.+区)", expand=True) # sample.value_counts() print(len(train_data_1)) print(len(test_data_1)) # 訓練データとテストデータに、エリア情報を追加 train_data_1["エリア"] = "" test_data_1["エリア"] = "" train_data_1["エリア"] = train_data_1["所在地"].str.extract("東京都((.+区.+?)[0-9]|(.+区.+?)[0-9]|(.+区.+))", expand=True) test_data_1["エリア"] = test_data_1["所在地"].str.extract("東京都((.+区.+?)[0-9]|(.+区.+?)[0-9]|(.+区.+))", expand=True) print(len(train_data_1)) print(len(test_data_1)) # 一部エリア情報に数値が残ってしまうので、消す train_data_1["エリア"] = train_data_1["エリア"].str.replace("[0-9]|[0-9]", "") train_data_1["エリア"] = train_data_1["エリア"].str.replace("一丁目", "") train_data_1["エリア"] = train_data_1["エリア"].str.replace("-", "") test_data_1["エリア"] = test_data_1["エリア"].str.replace("[0-9]|[0-9]", "") test_data_1["エリア"] = test_data_1["エリア"].str.replace("一丁目", "") test_data_1["エリア"] = test_data_1["エリア"].str.replace("二丁目", "") test_data_1["エリア"] = test_data_1["エリア"].str.replace("四丁目", "") print(len(train_data_1)) print(len(test_data_1)) # 坪単価情報の読込~結合 wards_value = pd.read_excel("23区_地域毎_坪単価.xlsx") train_data_1 = pd.merge(train_data_1, wards_value, on="エリア", how="left") test_data_1 = pd.merge(test_data_1, wards_value, on="エリア", how="left") train_data_1 = train_data_1.sort_values("id") test_data_1 = test_data_1.sort_values("id") train_data_1 = train_data_1.drop("Unnamed: 4", axis=1) test_data_1 = test_data_1.drop("Unnamed: 4", axis=1) # 地価の抜けデータを平均値で補間 # 江東区 train_data_1.loc[train_data_1["id"]==2767, "区"] = "江東区" train_data_1.loc[train_data_1["id"]==7559, "区"] = "江東区" train_data_1.loc[train_data_1["id"]==2767, "平均坪単価(万円)"] = 206.6 train_data_1.loc[train_data_1["id"]==7559, "平均坪単価(万円)"] = 206.6 # 港区 train_data_1.loc[train_data_1["id"]==17654, "区"] = "港区" train_data_1.loc[train_data_1["id"]==17654, "平均坪単価(万円)"] = 1321.4 print(len(train_data_1)) print(len(test_data_1)) print(train_data_1[train_data_1["平均坪単価(万円)"].isnull()]) print(test_data_1[test_data_1["平均坪単価(万円)"].isnull()])<jupyter_output>Empty DataFrame Columns: [id, アクセス, 所在地, 賃料, 築年数, 面積m2, 間取りID, 所在, 階層, 経度, 緯度, 部屋数, 1部屋当たり面積m2, 高さ率, 建物構造ID, 賃料/面積, エリア, 区, 地名, 平均坪単価(万円)] Index: [] Empty DataFrame Columns: [id, アクセス, 所在地, 築年数, 面積m2, 間取りID, 所在, 階層, 経度, 緯度, 部屋数, 1部屋当たり面積m2, 高さ率, 建物構造ID, エリア, 区, 地名, 平均坪単価(万円)] Index: [] <jupyter_text>### 2.13. 六本木駅からの距離情報を追加<jupyter_code># 六本木駅からの距離を追加 # 六本木駅 緯度: 35.662725 経度: 139.731217 from geopy import Point from geopy.distance import geodesic from geopy.distance import distance train_data_1["Roppongi_longtitude"] = 139.731217 train_data_1["Roppongi_latitude"] = 35.662725 train_data_1['point'] = train_data_1.apply(lambda row: Point(latitude=row['緯度'], longitude=row['経度']), axis=1) train_data_1['point_next'] = train_data_1.apply(lambda row: Point(latitude=row['Roppongi_latitude'], longitude=row['Roppongi_longtitude']), axis=1) test_data_1["Roppongi_longtitude"] = 139.731217 test_data_1["Roppongi_latitude"] = 35.662725 test_data_1['point'] = test_data_1.apply(lambda row: Point(latitude=row['緯度'], longitude=row['経度']), axis=1) test_data_1['point_next'] = test_data_1.apply(lambda row: Point(latitude=row['Roppongi_latitude'], longitude=row['Roppongi_longtitude']), axis=1) train_data_1['distance_km'] = train_data_1.apply(lambda row: distance(row['point'], row['point_next']).km if row['point_next'] is not None else float('nan'), axis=1) test_data_1['distance_km'] = test_data_1.apply(lambda row: distance(row['point'], row['point_next']).km if row['point_next'] is not None else float('nan'), axis=1) # 距離値を逆数にする Reciprocal train_data_1["Recciprocal_distance_km"] = (1 / train_data_1["distance_km"]).astype(float) test_data_1["Recciprocal_distance_km"] = (1 / test_data_1["distance_km"]).astype(float) # 不要なカラムを削除する train_data_1 = train_data_1.drop(["Roppongi_longtitude", "Roppongi_latitude", "point", "point_next", "distance_km"], axis=1) test_data_1 = test_data_1.drop(["Roppongi_longtitude", "Roppongi_latitude", "point", "point_next", "distance_km"], axis=1)<jupyter_output><empty_output><jupyter_text>### 2.14. 欠損値を区の平均値で埋める<jupyter_code>train_data_1["築年数"] = train_data_1.groupby('区')["築年数"].transform(lambda x:x.fillna(x.mean())) train_data_1["所在"] = train_data_1.groupby('区')["所在"].transform(lambda x:x.fillna(x.mean())) train_data_1["階層"] = train_data_1.groupby('区')["階層"].transform(lambda x:x.fillna(x.mean())) test_data_1["築年数"] = test_data_1.groupby('区')["築年数"].transform(lambda x:x.fillna(x.mean())) test_data_1["所在"] = test_data_1.groupby('区')["所在"].transform(lambda x:x.fillna(x.mean())) test_data_1["階層"] = test_data_1.groupby('区')["階層"].transform(lambda x:x.fillna(x.mean())) # 高さ率の更新 train_data_1["高さ率"] = (train_data_1["所在"] / train_data_1["階層"]).astype(float) test_data_1["高さ率"] = (test_data_1["所在"] / test_data_1["階層"]).astype(float)<jupyter_output><empty_output><jupyter_text>### 2.15. 区IDと区ごとの統計情報の追加<jupyter_code># 区にlabel encodingを適用 from sklearn.preprocessing import LabelEncoder le = LabelEncoder() le.fit(pd.concat([train_data_1["区"],test_data_1["区"]])) train_data_1["区ID"] = le.transform(train_data_1["区"]) test_data_1["区ID"] = le.transform(test_data_1["区"]) # 区ごとの統計情報の追加 # 例)新宿区のとある物件の築年数の統計値:(【1】-【2】)*【1】/【2】 # 【1】該当物件の築年数、【2】新宿区の物件の平均築年数:=(該当物件の築年数-)* # train_data_1 = train_data_1.assign(target_enc=train_data_1.groupby("区ID")["賃料/面積"].transform("mean").copy())<jupyter_output><empty_output><jupyter_text>### 2.16. 最寄りの路線ID、駅IDと、そこからの徒歩分数を追加<jupyter_code># 徒歩分数の整理のために初回のみ出力 # test_data_1["路線情報"] = test_data_1["アクセス"].str.extract("(\d{1,2}分)", expand=True) # train_data_1 = pd.concat([train_data_1, train_data_1['アクセス'].str.split('\t', expand=True)], axis=1).drop('アクセス', axis=1) # test_data_1 = pd.concat([test_data_1, test_data_1['アクセス'].str.split('\t', expand=True)], axis=1).drop('アクセス', axis=1) # train_access_1 = train_data_1[[0, 1, 2]] # train_access_2 = train_data_1[[4, 5, 6]] # train_access_3 = train_data_1[[8, 9, 10]] # test_access_1 = test_data_1[[0, 1, 2]] # test_access_2 = test_data_1[[4, 5, 6]] # test_access_3 = test_data_1[[8, 9, 10]] # train_access_1.to_excel("train_access_1.xlsx", header=True, index=False) # train_access_2.to_excel("train_access_2.xlsx", header=True, index=False) # train_access_3.to_excel("train_access_3.xlsx", header=True, index=False) # test_access_1.to_excel("test_access_1.xlsx", header=True, index=False) # test_access_2.to_excel("test_access_2.xlsx", header=True, index=False) # test_access_3.to_excel("test_access_3.xlsx", header=True, index=False) # 最寄り駅データの読み込み train_access_result = pd.read_excel("./アクセス/train_access_result.xlsx") test_access_result = pd.read_excel("./アクセス/test_access_result.xlsx") # データ結合 train_data_1 = pd.concat([train_data_1, train_access_result], axis=1) test_data_1 = pd.concat([test_data_1, test_access_result], axis=1) # 徒歩分数とfloat型に変換する train_data_1["徒歩分数"] = (train_data_1["徒歩分数"]).astype(float) test_data_1["徒歩分数"] = (test_data_1["徒歩分数"]).astype(float) # 路線名にlabel encodingを適用 le = LabelEncoder() le.fit(pd.concat([train_data_1["路線名"],test_data_1["路線名"]])) train_data_1["路線ID"] = le.transform(train_data_1["路線名"]) test_data_1["路線ID"] = le.transform(test_data_1["路線名"]) # 駅名にlabel encodingを適用 le = LabelEncoder() le.fit(pd.concat([train_data_1["駅名"],test_data_1["駅名"]])) train_data_1["駅ID"] = le.transform(train_data_1["駅名"]) test_data_1["駅ID"] = le.transform(test_data_1["駅名"])<jupyter_output><empty_output><jupyter_text>### 2.17. 重複物件の抽出・削除<jupyter_code># 重複確認用のデータ出力 # train_duplicate = pd.concat([train_data[["id", "賃料", "所在地"]], train_data_1[["間取りID", "面積m2", "所在", "階層"]]], axis=1) # test_duplicate = pd.concat([test_data[["id", "所在地"]], test_data_1[["間取りID", "面積m2", "所在", "階層"]]], axis=1) # 結果の出力 # train_duplicate.to_excel("train_duplicate.xlsx", header=True, index=False) # test_duplicate.to_excel("test_duplicate.xlsx", header=True, index=False) # 重複対象のデータ読み込み train_duplicate_flag = pd.read_excel("train_duplicate_flag.xlsx") test_duplicate_flag = pd.read_excel("test_duplicate_flag.xlsx") # データ数のチェック print("Train:", len(train_data_1), "-", len(train_duplicate_flag), "=", len(train_data_1)-len(train_duplicate_flag)) print("Test:", len(test_data_1), "-", len(test_duplicate_flag), "=", len(test_data_1)-len(test_duplicate_flag)) # 重複フラグを結合 train_data_1 = pd.merge(train_data_1, train_duplicate_flag, on="id") train_data_1 = train_data_1.rename(columns={"賃料_x":"賃料"}) train_data_1 = train_data_1.drop('賃料_y', axis=1) test_data_1 = pd.merge(test_data_1, test_duplicate_flag, on="id") # testデータからのみ、flag=1.0の列を削除 print("【削除前】", "train:", len(train_data_1), ", ", "test:", len(test_data_1)) #train_data_1 = train_data_1.loc[train_data_1["flag"] != 1.0] test_data_1 = test_data_1.loc[test_data_1["flag"] != 1.0] print("【削除後】", "train:", len(train_data_1), ", ", "test:", len(test_data_1)) # 不要なカラムを削除する train_data_1 = train_data_1.drop(["アクセス", "所在地", "エリア", "区", "No.", "地名", "路線名", "駅名"], axis=1) test_data_1 = test_data_1.drop(["アクセス", "所在地", "エリア", "区", "No.", "地名", "路線名", "駅名"], axis=1)<jupyter_output><empty_output><jupyter_text>## 3. データチェック<jupyter_code>train_data_1.isnull().sum() test_data_1.isnull().sum() len(train_data_1) len(train_data) len(test_data_1) len(test_data) train_data_1.head() test_data_1.head()<jupyter_output><empty_output><jupyter_text>## 4.学習 <jupyter_code># IDとflagの削除 train_data_1_no_ID = train_data_1.drop(["id", "flag"], axis=1) test_data_1_no_ID = test_data_1.drop(["id", "flag"], axis=1) # 特徴データと目的変数の設定 # train_x = train_data_1_no_ID.drop(["賃料", "賃料/面積"], axis=1) train_x = train_data_1_no_ID train_y = train_data_1_no_ID["賃料/面積"] test_x = test_data_1_no_ID import xgboost as xgb from xgboost import XGBClassifier from sklearn.metrics import log_loss, accuracy_score from sklearn.model_selection import KFold import pandas as pd from sklearn.decomposition import PCA import math from sklearn import model_selection from sklearn.ensemble import RandomForestRegressor # 2つの値から角度を求めて、配列を返す関数 def make_radian_row(pca_result): rad = [] for r in pca_result: rad.append(math.atan(r[0]/r[1])) return rad scores = [] # GBDT用のハイパーパラメータ params = {"objective": "reg:squarederror", "sileng":1, "random_state":71, "eval_metric":"rmse"} num_round = 50 # クロスバリデーション kf = KFold(n_splits=4, shuffle=True, random_state=71) for tr_idx, va_idx, in kf.split(train_x): tr_x, va_x = train_x.iloc[tr_idx], train_x.iloc[va_idx] tr_y, va_y = train_y.iloc[tr_idx], train_y.iloc[va_idx] # 訓練データを'区ID'でgroupbyを行い、'賃料/面積'の平均を求める。 tr_x['id_target_enc'] = tr_x.groupby('区ID')['賃料/面積'].transform('mean') # 検証データを使用しないで'区ID'カラムに対する'賃料/面積'の平均を求める va_x['id_target_enc'] = va_x['区ID'].map(tr_x.groupby('区ID')['賃料/面積'].mean()) test_x['id_target_enc'] = test_x['区ID'].map(tr_x.groupby('区ID')['賃料/面積'].mean()) # 検証データの行にtarget_encodingを入れる # train_x.loc[train_x.index[va_idx], 'item_target_enc'] = va_x['item_target_enc'] # tr_x, va_xより、"賃料", "賃料/面積"を削除 tr_x = tr_x.drop(["賃料", "賃料/面積"], axis=1) va_x = va_x.drop(["賃料", "賃料/面積"], axis=1) # 主成分分析 pca = PCA(n_components=2) pca.fit(tr_x) # 角度データの追加 tr_x["rad"] = make_radian_row(pca.transform(tr_x)) va_x["rad"] = make_radian_row(pca.transform(va_x)) test_x["rad"] = make_radian_row(pca.transform(test_x)) # GBDTで学習実行 dtrain = xgb.DMatrix(tr_x, label=tr_y) # enable_categorical=True dvalid = xgb.DMatrix(va_x, label=va_y) watchlist = [(dtrain, "train"), (dvalid, "eval")] model = xgb.train(params, dtrain, num_round, evals=watchlist) tr_x = tr_x.drop("rad", axis=1) va_x = va_x.drop("rad", axis=1) # import xgboost as xgb # from xgboost import XGBClassifier # from sklearn.metrics import log_loss, accuracy_score # from sklearn.model_selection import KFold # scores = [] # # GBDT用のハイパーパラメータ # params = {"objective": "reg:squarederror", "sileng":1, "random_state":71, "eval_metric":"rmse"} # num_round = 50 # # クロスバリデーション # kf = KFold(n_splits=4, shuffle=True, random_state=71) # for tr_idx, va_idx, in kf.split(train_x): # tr_x, va_x = train_x.iloc[tr_idx], train_x.iloc[va_idx] # tr_y, va_y = train_y.iloc[tr_idx], train_y.iloc[va_idx] # # GBDTで学習実行 # dtrain = xgb.DMatrix(tr_x, label=tr_y) # enable_categorical=True # dvalid = xgb.DMatrix(va_x, label=va_y) # dtest = xgb.DMatrix(test_x) # watchlist = [(dtrain, "train"), (dvalid, "eval")] # model = xgb.train(params, dtrain, num_round, evals=watchlist)<jupyter_output><empty_output><jupyter_text>## 5. 検証<jupyter_code># 特徴量の予測結果への貢献度(Fスコア)を可視化 import matplotlib.pyplot as plt import matplotlib import numpy as np font = {'family' : 'Meiryo'} # matplotlibのデフォルトフォントをTakaoGothicに設定 plt.rc('font', **font) xgb.plot_importance(model, importance_type = "gain") plt.show() # ヒストグラムと散布図の行列を作成 from pandas.plotting import scatter_matrix x = scatter_matrix(train_data_1_no_ID, alpha=1, figsize=(20, 10), diagonal='hist') # PCA分析結果を可視化する transformed = pca.fit_transform(tr_x) # 主成分をプロットする plt.subplot(1, 2, 2) plt.scatter(transformed[:, 0], transformed[:, 1]) plt.title('principal component') plt.xlabel('pc1') plt.ylabel('pc2') # 主成分の次元ごとの寄与率を出力する print(pca.explained_variance_ratio_) # グラフを表示する plt.show()<jupyter_output>[0.98837778 0.01086506] <jupyter_text>## 6. 予測<jupyter_code>tr_x.head() test_x.head() # テストデータでの予測 # 角度データの追加 test_x["rad"] = make_radian_row(pca.transform(test_x)) # # PCAで出力した角度情報を元データに結合 # test_x = pd.concat([test_x, test_x_pca["rad"]], axis=1) dtest = xgb.DMatrix(test_x) pred = model.predict(dtest) pred len(pred) len(test_data_1)<jupyter_output><empty_output><jupyter_text>## 7. データ出力<jupyter_code># 予測結果の結合 test_data_1["賃料/面積"] = pred # 予測結果から賃料を算出し、int型に変換 test_data_1["賃料"] = (test_data_1["面積m2"] * test_data_1["賃料/面積"]).astype(int) result = test_data_1[["id", "賃料"]] result len(result) # 除外していた重複idの賃料を戻す test_duplicate_rentvalue = pd.read_excel("test_duplicate_rentvalue.xlsx") result = pd.concat([result, test_duplicate_rentvalue]) result = result.sort_values("id") len(result) result result.to_csv("result.csv", header=False, index=False)<jupyter_output><empty_output>
no_license
/00_Python/SIGNATE/20210818_【SOTA】マイナビ × SIGNATE Student Cup 2019 賃貸物件の家賃予測/.ipynb_checkpoints/RentForcast_210926_01-checkpoint.ipynb
Kosuke-Yanai/KY
23
<jupyter_start><jupyter_text># Learning Together - Data Analysis$e^{i\pi} + 1 = 0$<jupyter_code>import numpy as np import pandas as pd from pandas import Series, DataFrame url = 'http://archive.ics.uci.edu/ml/machine-learning-databases/00484/' dframe_trip = pd.read_csv('tripadvisor_review.csv', sep = ',') dframe_trip.head(10) dframe_trip.tail(10) dframe_trip['Category 7'].mean() def max_to_min(arr): return arr.max() - arr.min() trip = dframe_trip.groupby('Category 10') trip.describe() trip.agg(max_to_min) #Fungsi agregasi yang menghitung nilai max dari setiap kolom dan mengurangi nilai min dari kolom trip.agg('mean') dframe_trip['Cat 10/4 Ratio'] = dframe_trip['Category 10'] / dframe_trip['Category 4'] dframe_trip.head() #instead of groupby we can use pivot table dframe_trip.pivot_table(index=['Category 10']) %matplotlib inline dframe_trip.plot(kind='scatter', x='Category 7', y='Category 10')<jupyter_output><empty_output>
permissive
/.ipynb_checkpoints/Aggregation-checkpoint.ipynb
buildGather/learnDataAnalysis
1
<jupyter_start><jupyter_text>--- _You are currently looking at **version 1.0** of this notebook. To download notebooks and datafiles, as well as get help on Jupyter notebooks in the Coursera platform, visit the [Jupyter Notebook FAQ](https://www.coursera.org/learn/python-text-mining/resources/d9pwm) course resource._ ---*Note: Some of the cells in this notebook are computationally expensive. To reduce runtime, this notebook is using a subset of the data.*# Case Study: Sentiment Analysis### Data Prep<jupyter_code>import pandas as pd import numpy as np # Read in the data df = pd.read_csv('Amazon_Unlocked_Mobile.csv') # Sample the data to speed up computation # Comment out this line to match with lecture df = df.sample(frac=0.1, random_state=10) df.head() # Drop missing values df.dropna(inplace=True) # Remove any 'neutral' ratings equal to 3 df = df[df['Rating'] != 3] # Encode 4s and 5s as 1 (rated positively) # Encode 1s and 2s as 0 (rated poorly) df['Positively Rated'] = np.where(df['Rating'] > 3, 1, 0) df.head(10) # Most ratings are positive df['Positively Rated'].mean() from sklearn.model_selection import train_test_split # Split data into training and test sets X_train, X_test, y_train, y_test = train_test_split(df['Reviews'], df['Positively Rated'], random_state=0) print('X_train first entry:\n\n', X_train.iloc[0]) print('\n\nX_train shape: ', X_train.shape) <jupyter_output>X_train first entry: Everything about it is awesome! X_train shape: (23052,) <jupyter_text># CountVectorizer<jupyter_code>from sklearn.feature_extraction.text import CountVectorizer # Fit the CountVectorizer to the training data vect = CountVectorizer().fit(X_train) vect.get_feature_names()[::2000] len(vect.get_feature_names()) # transform the documents in the training data to a document-term matrix X_train_vectorized = vect.transform(X_train) X_train_vectorized from sklearn.linear_model import LogisticRegression # Train the model model = LogisticRegression() model.fit(X_train_vectorized, y_train) from sklearn.metrics import roc_auc_score # Predict the transformed test documents predictions = model.predict(vect.transform(X_test)) print('AUC: ', roc_auc_score(y_test, predictions)) # get the feature names as numpy array feature_names = np.array(vect.get_feature_names()) # Sort the coefficients from the model sorted_coef_index = model.coef_[0].argsort() # Find the 10 smallest and 10 largest coefficients # The 10 largest coefficients are being indexed using [:-11:-1] # so the list returned is in order of largest to smallest print('Smallest Coefs:\n{}\n'.format(feature_names[sorted_coef_index[:10]])) print('Largest Coefs: \n{}'.format(feature_names[sorted_coef_index[:-11:-1]]))<jupyter_output>Smallest Coefs: ['worst' 'terrible' 'slow' 'junk' 'waste' 'sucks' 'poor' 'useless' 'disappointed' 'horrible'] Largest Coefs: ['excelent' 'excellent' 'excelente' 'perfectly' 'love' 'perfect' 'exactly' 'great' 'best' 'awesome'] <jupyter_text># Tfidf<jupyter_code>from sklearn.feature_extraction.text import TfidfVectorizer # Fit the TfidfVectorizer to the training data specifiying a minimum document frequency of 5 vect = TfidfVectorizer(min_df=5).fit(X_train) len(vect.get_feature_names()) X_train_vectorized = vect.transform(X_train) model = LogisticRegression() model.fit(X_train_vectorized, y_train) predictions = model.predict(vect.transform(X_test)) print('AUC: ', roc_auc_score(y_test, predictions)) feature_names = np.array(vect.get_feature_names()) sorted_tfidf_index = X_train_vectorized.max(0).toarray()[0].argsort() print('Smallest tfidf:\n{}\n'.format(feature_names[sorted_tfidf_index[:10]])) print('Largest tfidf: \n{}'.format(feature_names[sorted_tfidf_index[:-11:-1]])) sorted_coef_index = model.coef_[0].argsort() print('Smallest Coefs:\n{}\n'.format(feature_names[sorted_coef_index[:10]])) print('Largest Coefs: \n{}'.format(feature_names[sorted_coef_index[:-11:-1]])) # These reviews are treated the same by our current model print(model.predict(vect.transform(['not an issue, phone is working', 'an issue, phone is not working'])))<jupyter_output>[0 0] <jupyter_text># n-grams<jupyter_code># Fit the CountVectorizer to the training data specifiying a minimum # document frequency of 5 and extracting 1-grams and 2-grams vect = CountVectorizer(min_df=5, ngram_range=(1,2)).fit(X_train) X_train_vectorized = vect.transform(X_train) len(vect.get_feature_names()) model = LogisticRegression() model.fit(X_train_vectorized, y_train) predictions = model.predict(vect.transform(X_test)) print('AUC: ', roc_auc_score(y_test, predictions)) feature_names = np.array(vect.get_feature_names()) sorted_coef_index = model.coef_[0].argsort() print('Smallest Coefs:\n{}\n'.format(feature_names[sorted_coef_index[:10]])) print('Largest Coefs: \n{}'.format(feature_names[sorted_coef_index[:-11:-1]])) # These reviews are now correctly identified print(model.predict(vect.transform(['not an issue, phone is working', 'an issue, phone is not working'])))<jupyter_output>[1 0]
no_license
/CURSO4/semana3/Case+Study+-+Sentiment+Analysis.ipynb
sandrarairan/Ciencias-Datos-Aplicada-Python-coursera
4
<jupyter_start><jupyter_text># K Nearest Neighbors Project <jupyter_code>import pandas as pd import numpy as np import matplotlib.pyplot as plt import seaborn as sns %matplotlib inline sns.set()<jupyter_output><empty_output><jupyter_text>## Get the Data<jupyter_code>df = pd.read_csv('KNN_Project_Data')<jupyter_output><empty_output><jupyter_text>**Check the head of the dataframe.**<jupyter_code>df.head()<jupyter_output><empty_output><jupyter_text># Exploratory Data Analysis (EDA)<jupyter_code>sns.pairplot(df, hue='TARGET CLASS', palette='coolwarm')<jupyter_output><empty_output><jupyter_text># Standardize the Variables<jupyter_code>from sklearn.preprocessing import StandardScaler scaler = StandardScaler() scaler.fit(df.drop('TARGET CLASS', axis=1)) # Compute the mean and std to be used for later scaling. scaled_features = scaler.transform(df.drop('TARGET CLASS', axis=1)) # Perform standardization by centering and scaling df_feat = pd.DataFrame(scaled_features, columns=df.columns[:-1]) df_feat.head(3)<jupyter_output><empty_output><jupyter_text># Train Test Split<jupyter_code>from sklearn.model_selection import train_test_split X = df_feat y = df['TARGET CLASS'] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=101)<jupyter_output><empty_output><jupyter_text># Using KNN<jupyter_code>from sklearn.neighbors import KNeighborsClassifier knn = KNeighborsClassifier(n_neighbors=1) knn.fit(X_train, y_train)<jupyter_output><empty_output><jupyter_text># Predictions and Evaluations<jupyter_code>pred = knn.predict(X_test) from sklearn.metrics import classification_report, confusion_matrix print(confusion_matrix(y_test, pred)) print(classification_report(y_test, pred))<jupyter_output> precision recall f1-score support 0 0.73 0.72 0.72 152 1 0.71 0.72 0.72 148 accuracy 0.72 300 macro avg 0.72 0.72 0.72 300 weighted avg 0.72 0.72 0.72 300 <jupyter_text># Choosing a K Value<jupyter_code># Trying results on different k values error_rate = [] for i in range(1, 40): knn = KNeighborsClassifier(n_neighbors=i) knn.fit(X_train, y_train) pred_i = knn.predict(X_test) error_rate.append(np.mean(pred_i != y_test)) plt.figure(figsize=(10, 6)) # X-axis, Y-asix plt.plot(range(1, 40), error_rate, color="blue", linestyle='--', marker='o', markerfacecolor='red', markersize=10) plt.title('Error Rate vs. K Value') plt.xlabel('K') plt.ylabel('Error Rate')<jupyter_output><empty_output><jupyter_text>## Retrain with new K Value<jupyter_code>knn = KNeighborsClassifier(n_neighbors=30) knn.fit(X_train, y_train) # give input and output to learn pattren pred = knn.predict(X_test) # give input to model to predict output value itself print(confusion_matrix(y_test, pred)) # now compare predictied values and actual values print('\n') print(classification_report(y_test, pred)) knn = KNeighborsClassifier(n_neighbors=31) knn.fit(X_train, y_train) # give input and output to learn pattren pred = knn.predict(X_test) # give input to model to predict output value itself print(confusion_matrix(y_test, pred)) # now compare predictied values and actual values print('\n') print(classification_report(y_test, pred))<jupyter_output>[[123 29] [ 19 129]] precision recall f1-score support 0 0.87 0.81 0.84 152 1 0.82 0.87 0.84 148 accuracy 0.84 300 macro avg 0.84 0.84 0.84 300 weighted avg 0.84 0.84 0.84 300
no_license
/K Nearest Neighbors Project.ipynb
hunain-saeed/KNN-pdsml
10
<jupyter_start><jupyter_text><jupyter_code>import pandas as pd df = pd.read_csv('drug review processed.csv.gz') df.info() df.head() df = df.drop(columns=df.columns[0]) df.head() df.groupby('vaderSentimentLabel').size() import matplotlib.pyplot as plt df.groupby('vaderSentimentLabel').count().plot.bar() plt.show() df.groupby('ratingSentimentLabel').size() df.groupby('ratingSentimentLabel').count().plot.bar() plt.show() df.groupby('ratingSentiment').size() positive_vader_sentiments = df[df.ratingSentiment == 2] positive_string = [] for s in positive_vader_sentiments.cleanReview: positive_string.append(s) positive_string = pd.Series(positive_string).str.cat(sep=' ') from wordcloud import WordCloud wordcloud = WordCloud(width=2000,height=1000,max_font_size=200).generate(positive_string) plt.imshow(wordcloud,interpolation='bilinear') plt.show() for s in positive_vader_sentiments.cleanReview[:20]: if 'side effect' in s: print(s) negative_vader_sentiments = df[df.ratingSentiment == 1] negative_string = [] for s in negative_vader_sentiments.cleanReview: negative_string.append(s) negative_string = pd.Series(negative_string).str.cat(sep=' ') from wordcloud import WordCloud wordcloud = WordCloud(width=2000,height=1000,max_font_size=200).generate(negative_string) plt.imshow(wordcloud,interpolation='bilinear') plt.axis('off') plt.show() for s in negative_vader_sentiments.cleanReview[:20]: if 'side effect' in s: print(s) neutral_vader_sentiments = df[df.ratingSentiment == 0] neutral_string = [] for s in neutral_vader_sentiments.cleanReview: neutral_string.append(s) neutral_string = pd.Series(neutral_string).str.cat(sep=' ') from wordcloud import WordCloud wordcloud = WordCloud(width=2000,height=1000,max_font_size=200).generate(neutral_string) plt.imshow(wordcloud,interpolation='bilinear') plt.axis('off') plt.show() for s in neutral_vader_sentiments.cleanReview[:20]: if 'side effect' in s: print(s) from sklearn.feature_extraction.text import TfidfVectorizer tfidf = TfidfVectorizer(stop_words='english',ngram_range=(1,2)) features = tfidf.fit_transform(df.cleanReview) labels = df.vaderSentiment features.shape from sklearn.model_selection import train_test_split from sklearn.feature_extraction.text import TfidfTransformer from sklearn.naive_bayes import MultinomialNB x_train,x_test,y_train,y_test = train_test_split(df['cleanReview'],df['ratingSentimentLabel'],random_state=0) from sklearn.linear_model import LogisticRegression from sklearn.ensemble import RandomForestClassifier from sklearn.svm import LinearSVC from sklearn.neighbors import KNeighborsClassifier from sklearn.model_selection import cross_val_score from sklearn.preprocessing import StandardScaler models = [RandomForestClassifier(n_estimators=200,max_depth=3,random_state=0),LinearSVC(),MultinomialNB(),LogisticRegression(random_state=0,solver='lbfgs',max_iter=2000,multi_class='auto')] CV = 5 cv_df = pd.DataFrame(index=range(CV * len(models))) entries = [] for model in models: model_name = model.__class__.__name__ accuracies = cross_val_score(model,features,labels,scoring='accuracy',cv=CV) for fold_idx,accuracy in enumerate(accuracies): entries.append((model_name,fold_idx,accuracy)) cv_df = pd.DataFrame(entries,columns=['model_name','fold_idx','accuracy']) cv_df cv_df.groupby('model_name').accuracy.mean() from sklearn.preprocessing import Normalizer model = LinearSVC('l2') x_train,x_test,y_train,y_test = train_test_split(features,labels,test_size=0.25,random_state=0) normalize = Normalizer() x_train = normalize.fit_transform(x_train) x_test = normalize.transform(x_test) model.fit(x_train,y_train) y_pred = model.predict(x_test) from sklearn.metrics import accuracy_score print(accuracy_score(y_test, y_pred)) from sklearn.metrics import confusion_matrix conf_mat = confusion_matrix(y_test,y_pred) conf_mat from mlxtend.plotting import plot_confusion_matrix fig,ax = plot_confusion_matrix(conf_mat=conf_mat,colorbar=True,show_absolute=True,cmap='viridis') from sklearn.metrics import classification_report print(classification_report(y_test,y_pred,target_names= df['ratingSentimentLabel'].unique())) #Precision is ratio between all adequately identified errors and all identified errors #Positive precision means whenever positive was predicted, the model was right 94% of time #Neutral precision means whenever neutral was predicted, the model was right 93% of time #Negative precision means whenever negative was predicted the model was right 93% of time # With a weighted average of 93% This model is quite good at recognizing error #Recall or sensitivity or True Positive Rate is a ratio of all identified errors and all existing errors #Positive recall means positive was selected 63% of the times it should have been selected #Neutral recall means neutral was selected 95% of the times it should have been selected #Negative recall means negative was selected 94% of the times it should have been selected # with weighted average of 93%. this model is quite good at classifying error,it seem to be weaker in classifying positive class with recall percentage of 63% <jupyter_output><empty_output>
no_license
/Opinion_Mining_using_the_UCI_Drug_Review_Data_(Part_2)_Sentiment_Prediction_Using_Machine_Learning_Classification_Algorithms(Scikit_Learn_Implementatition).ipynb
zainabnatiq/Opinion-Mining-using-the-UCI-Drug-Review-Dataset
1
<jupyter_start><jupyter_text><jupyter_code><jupyter_output><empty_output><jupyter_text># Introduction to Regression with Neural Networks in Tensorflow There are many definition for a regression problem but in our case, we're going to simplify it: predicting a numerical variable based on some other combination of variables, even shorter...predicting a number<jupyter_code># Import TensorFlow import tensorflow as tf print(tf.__version__)<jupyter_output>2.4.1 <jupyter_text>## Creating data to view and fit<jupyter_code>import numpy as np import matplotlib.pyplot as plt # Create features X = np.array([-7.0, -4.0, -1.0, 2.0, 5.0, 8.0, 11.0, 14.0]) # Create labels y = np.array([3.0,6.0,9.0,12.0,15.0,18.0,21.0,24.0]) # Visualize it plt.scatter(X,y) y == X + 10<jupyter_output><empty_output><jupyter_text>## Input and output shapes<jupyter_code># Create a demo tensor for our housing price prediction problem house_info = tf.constant(["bedroom", "bathroom", "garage"]) house_price = tf.constant([939700]) house_info, house_price X[0], y[0] X[1], y[1] input_shape = X[0].shape output_shape = y[0].shape input_shape, output_shape X[0].ndim # this is why it has no shape X[0], y[0] #we want a model that can take an input X and output y, use 1 x value to predict 1 y value # Turn our Numpy array into tensors X = tf.cast(tf.constant(X), dtype = tf.float32) y = tf.cast(tf.constant(y), dtype = tf.float32) X, y input_shape = X[0].shape output_shape = y[0].shape input_shape, output_shape<jupyter_output><empty_output><jupyter_text>## Steps in modelling with TensorFlow 1. **Creating a model** - define the input and output layers, as well as the hidden layers of a deep learning model. 2. **Compiling a model ** - define the loss function (in other words, the function which tells our model how wrong it is) and the optimizer (tells our model how to imporve the patterns its learning) and evaluation metrics (what we can use to interpret the moerformance of our model). 3. ** Fitting a model** - letting the model try to find patterns between X and y (features and labels).<jupyter_code>tf.random.set_seed(42) # 1. Create a model using the Sequential API model = tf.keras.Sequential([ tf.keras.layers.Dense(1) ]) #2. Compile the model model.compile(loss = tf.keras.losses.mae, optimizer = tf.keras.optimizers.SGD(), metrics = ["mae"]) # 3. Fit the model model.fit(X, y, epochs = 100) # check out x and y X, y # try and make a prediction y_pred = model.predict([17.0]) y_pred<jupyter_output><empty_output><jupyter_text>## Imporoving our model We can improve our model, by altering the steps we took to create a model. 1. **Creating a model** - here we might add more layers, increase the number of hidden units (all called neurons) within each of the hidden layers, change the activation function of each layer. 2. **Compiling a model** - here we might change the optimization function or perhaps the learning rate of the optimization function 3. **Fitting a model** - here we might fit a model for more epochs (leave it training for longer or on more data)<jupyter_code># :et's rebild our model import tensorflow as tf from tensorflow import keras # 1. Create the model model = tf.keras.Sequential([ tf.keras.layers.Dense(1) ]) #2. Compile the model model.compile(loss = tf.keras.losses.mae, optimizer = tf.keras.optimizers.SGD(), metrics = ['mae']) #3. Fit the model model.fit(X, y, epochs = 100) # Let's see if model's prediction has imprtoved model.predict([17.0]) # Let's rebuild the model model = tf.keras.Sequential([ tf.keras.layers.Dense(100, activation = 'relu'), tf.keras.layers.Dense(100, activation = 'relu'), tf.keras.layers.Dense(1) ]) model.compile(loss = tf.keras.losses.mae, optimizer = tf.keras.optimizers.Adam(), metrics = ['mae']) model.fit(X, y, epochs = 500) model.predict([17.0])<jupyter_output><empty_output><jupyter_text>## Evaluating a model<jupyter_code><jupyter_output><empty_output>
permissive
/neural_network_regression_with_tensorflow.ipynb
artms-18/tensorflow_fundamentals
7
<jupyter_start><jupyter_text> ## 2. Linear regression with multiple variables In this part, you will implement linear regression with multiple variables to predict the prices of houses. You want to predict the price of a house given its size and number of bedrooms. To achieve this, you need to train a model on data collected on housing prices given its size and number of bedrooms. The file ex1data2.txt contains a training set of housing prices in Portland, Oregon. The first column is the size of the house (in square feet), the second column is the number of bedrooms, and the third column is the price of the house. <jupyter_code>import numpy as np import pandas as pd import matplotlib.pyplot as plt # Load Data data = pd.read_csv('data/ex1data2.txt', names=['size','bedroom','price']); X = data.iloc[:, :2].values; y = data.iloc[:, 2].values; m = len(y); data.head() data.shape def h_of_x(X, theta): return np.dot(X,theta) def computeCostMulti(X, y, theta): m = len(y) J = (np.sum((np.dot(X,theta) - y)**2))/(2*m) return J<jupyter_output><empty_output><jupyter_text>### 2.1 Feature Normalization <jupyter_code>def featureNormalize(X): X_norm = X.copy() mu = np.zeros(X.shape[1]) sigma = np.zeros(X.shape[1]) #====================== mu = np.mean(X,axis=0) sigma = np.std(X,axis=0) X_norm = np.divide((X - mu),sigma) #==================== return X_norm, mu, sigma # Scale features and set them to zero mean print ('Normalizing Features ...') X, mu, sigma = featureNormalize(X) print ('[mu] [sigma]') print (mu, sigma) # Add intercept term to X X = np.concatenate((np.ones((m, 1)), X), axis=1)<jupyter_output>Normalizing Features ... [mu] [sigma] [2000.68085106 3.17021277] [7.86202619e+02 7.52842809e-01] <jupyter_text>### 2.2 Gradient Descent **Implementation Note:** In the multivariate case, the cost function can also be written in the following vectorized form: $$J(\theta) = \frac{1}{2m}(X\theta-\vec{y})^T (X\theta-\vec{y})$$ $$X = \begin{bmatrix} ....(x^{(1)})^T....\\ ....(x^{(2)})^T....\\ . \\ . \\ . \\ ....(x^{(m)})^T.... \end{bmatrix}, \ \ \ \ \ \ \ \ \ \vec{y} = \begin{bmatrix} y^{(1)}\\ y^{(2)}\\ .\\ . \\ .\\ y^{(m)} \end{bmatrix}$$ <jupyter_code>def gradientDescentMulti(X, y, theta, alpha, num_iters): m = len(y); # number of training examples J_history = np.zeros([num_iters, 1]); for iter in range(num_iters): #============ Your code here ============= #theta = theta - (alpha/m)*np.sum((h_of_x(X,theta)-y)[:,None]*X,axis=0) theta = theta - (alpha/m)*(np.dot(X.T,(h_of_x(X,theta)-y))) #================================== J_history[iter] = computeCostMulti(X, y, theta) return theta, J_history print(y.shape) print(h_of_x(X,np.zeros(3)).shape) # Choose some alpha value alpha = 0.01 num_iters = 400 def gd(alpha): print ('Running gradient descent ...') # Init Theta and Run Gradient Descent thet = np.zeros(3) thet, J_history = gradientDescentMulti(X, y, thet, alpha, num_iters) return thet, J_history thet, J_history = gd(alpha)<jupyter_output>Running gradient descent ... <jupyter_text>### 2.3 Selecting learning rates In this part of the exercise, you will get to try out different learning rates for the dataset and find a learning rate that converges quickly. We rec- ommend trying values of the learning rate $\alpha$ on a log-scale, at multiplicative steps of about 3 times the previous value (i.e., 0.3, 0.1, 0.03, 0.01 and so on). You may also want to adjust the number of iterations you are running if that will help you see the overall trend in the curve.<jupyter_code># Choose some alpha value alphas = [0.03,0.01,1,0.3] thetas = [] num_iters = 400 def costs(): J = [] for i in alphas: alpha = i thet, j = gd(i) J.append(j) thetas.append(thet) return J J_hists = costs() axs = [] fig = plt.figure(figsize=(15,6)) for i in range(len(J_hists)): axs.append(fig.add_subplot('22'+str(i))) axs[-1].plot(J_hists[i]) axs[-1].set_xlabel('Number of iterations') axs[-1].set_ylabel('Cost J') #axs[-1].title = str(alphas[i]) plt.show <jupyter_output>Running gradient descent ... Running gradient descent ... Running gradient descent ... Running gradient descent ... <jupyter_text>Now make a prediction for a house with 1650 square feet and 3 bedrooms<jupyter_code>print ('Theta computed from gradient descent: ') print (thet) price = None # Estimate the price of a 1650 sq-ft, 3 br house ####### START CODE ########### example = np.array([[1, 1650, 3]]) # the test example example_norm = np.ones(example.shape) # a variable is initialized with an array of ones of same shape as the test example # this variable will hold the normalized features example_norm.dtype = np.float # Note that the array datatype must be float32 and not int # apply feature scaling and normalization example_norm[:, 1:] = np.divide((example[:, 1:] - mu),sigma) # Now make predictions price = h_of_x(example_norm, thet) ####### END CODE ############ print ('Predicted price of a 1650 sq-ft, 3 br house') print ('using gradient descent: ') print (price) theta[0]*examp example_norm[:, 1:]<jupyter_output><empty_output><jupyter_text>### 3.3 Normal Equations $$\theta = (X^TX)^{-1}X^T\vec{y}$$ Using this formula does not require any feature scaling, and you will get an exact solution in one calculation: there is no \loop until convergence" like in gradient descent. Complete the code in normaleqn.py to use the formula above to calcu- late $\theta$. Remember that while you don't need to scale your features, we still need to add a column of 1's to the X matrix to have an intercept term<jupyter_code>def normalEqn(X,y): return np.dot((np.linalg.inv(np.dot(X.T,X))),np.dot(X.T,y)) print ('Solving with normal equations...') # Calculate the parameters from the normal equation theta = normalEqn(X, y) # Display normal equation's result print ('Theta computed from the normal equations:') print (' %s \n' % theta)<jupyter_output>Solving with normal equations... Theta computed from the normal equations: [340412.65957447 109447.79646964 -6578.35485416] <jupyter_text>Now, since you have found $\theta$ using normal equation method, use it to make a price prediction for a 1650-square-foot house with 3 bedrooms. You should find that gives the same predicted price as the value you obtained using the model fit with gradient descent.<jupyter_code>price = None # Estimate the price of a 1650 sq-ft, 3 br house ####### START CODE ########### ####### END CODE ############ print ('Predicted price of a 1650 sq-ft, 3 br house') print ('using gradient descent: ') print (price)<jupyter_output>Predicted price of a 1650 sq-ft, 3 br house using gradient descent: None
no_license
/Beginners/Week3/codelab/Linear regression with multiple variables.ipynb
Otsebolu/cycle2-resource-materials
7
<jupyter_start><jupyter_text># Preview one original file<jupyter_code>filename = folder+'TRABEAE12903CDD8EE.mp3' orig, sr = librosa.load(filename, sr=16000) IPython.display.Audio(orig, rate=sr)<jupyter_output><empty_output><jupyter_text># Sonify mel-spectrogram of that example<jupyter_code>recon, sr = transform_and_restore(filename, sr=16000, n_fft=512, n_mels=48, hop_length=256, hop_stride=10, hop_fill=True, plot_mels=True) IPython.display.Audio(recon, rate=sr)<jupyter_output><empty_output><jupyter_text># Generate all the audio examples<jupyter_code>mels = [128, 96, 48, 32, 24, 16, 8] temp = [1, 2, 3, 4, 5, 10] srs = [12, 16] files = [] for r, d, f in os.walk(folder): for file in f: if '.mp3' in file: for n_mel in mels: for tmp in temp: for sr in srs: recon, o_sr = transform_and_restore(os.path.join(folder, file), sr=sr*1000, n_mels=n_mel, hop_stride=tmp, hop_fill=True) new_file_directory = os.path.join(folder, file.replace('.mp3','')) if not os.path.exists(new_file_directory): os.makedirs(new_file_directory) new_filename = '{}/{}k-mel{}-x{}'.format(new_file_directory, sr, n_mel , tmp) librosa.output.write_wav(new_filename+'.wav', recon, o_sr) # Convert wav to flac ! ffmpeg -i {new_filename}.wav -vn -ar 16000 -sample_fmt s16 -ss 0 -t 30 {new_filename}.flac os.remove(new_filename+'.wav')<jupyter_output><empty_output>
non_permissive
/Sonify.ipynb
andrebola/ICASSP2020
3
<jupyter_start><jupyter_text># MEE - AQUISIÇÃO DE DADOSPrimeira tentativa usando o celular do Matheus com n=1 e n=9. Foi aqui que percebemos que não estamos considerando todas as incertezas. ERRO NORMALIZADO INCOMPATÍVEL.## Importando Pacotes<jupyter_code>import matplotlib.pyplot as plt from numpy import array, absolute, sqrt<jupyter_output><empty_output><jupyter_text>## Declarando os Arrays### n=1<jupyter_code>encoder1 = array([000.0000, 010.1250, 020.0249, 030.1499, 040.0499, 050.4000, 060.2999, 070.1999, 080.0999, 090.0000]) app1 = array([1.86371, 10.28004, 20.11119, 29.67642, 39.47946, 49.90371, 59.80443, 70.37051, 81.15369, 90.8339]) ictzApp1 = array([0.00245, 0.00472, 0.15142, 0.02835, 0.04445, 0.12088, 0.10862, 0.0249, 0.66702, 0.06263]) ictzEnc1 = array([0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225])<jupyter_output><empty_output><jupyter_text>### n=9<jupyter_code>encoder9 = array([000.0000, 009.8999, 020.2500, 030.1600, 040.2750, 049.9500, 060.0750, 070.8750, 080.3249, 090.0000]) app9 = array([2.08507, 9.61817, 20.3044, 29.7121, 38.96702, 48.85498, 59.3227, 70.47999, 81.40712, 90.16033]) ictzApp9 = array([0.01399, 0.03532, 0.06421, 0.13157, 0.15539, 0.18738, 0.17565, 0.33738, 0.10946, 0.2565]) ictzEnc9 = array([0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225, 0.225])<jupyter_output><empty_output><jupyter_text>## Aparência dos Gráficos<jupyter_code>#DEFININDO ALGUNS PARÂMETROS DO GRÁFICO %matplotlib inline %config InlineBackend.figure_format = 'svg' params = {'figure.figsize': [10, 5], 'axes.labelsize': 15, 'axes.titlesize':19, 'font.size': 13, 'legend.fontsize': 13, 'xtick.labelsize': 11, 'ytick.labelsize': 11, 'axes.axisbelow': True } plt.rcParams.update(params)<jupyter_output><empty_output><jupyter_text>## Plotando### n=1<jupyter_code>#PLOTANDO A IDEAL plt.scatter(encoder1, encoder1, s=100, alpha=0.7, label="IDEAL", color="crimson") #PLOTANDO A COMPARAÇÃO plt.scatter(encoder1, app1, s=70, label="OBTIDO", color="darkblue") #OUTROS PARÂMETROS DO GRÁFICO plt.legend() plt.title('Comparação com n=1') plt.xlabel('Ângulo do Encoder') plt.ylabel('Ângulo do Aplicativo') plt.xlim(0, 90) plt.ylim(0, 90) plt.xticks(array([0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]), ['0°', '10°', '20°', '30°', '40°', '50°', '60°', '70°', '80°', '90°']) plt.yticks(array([0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]), ['0°', '10°', '20°', '30°', '40°', '50°', '60°', '70°', '80°', '90°']) plt.grid(True) plt.savefig('PDF/celularMatheusN1.pdf', format='pdf', bbox_inches = 'tight') plt.savefig('EPS/celularMatheusN1.eps', format='eps', bbox_inches = 'tight') #EXIBINDO plt.show() #PLOTANDO A IDEAL plt.errorbar(encoder1, encoder1, yerr=ictzEnc1, alpha=0.7, label="IDEAL", color="crimson", fmt='o') #PLOTANDO A COMPARAÇÃO plt.errorbar(encoder1, app1, yerr=ictzApp1, label="OBTIDO", color="darkblue", fmt='o') #OUTROS PARÂMETROS DO GRÁFICO plt.legend() plt.title('Ampliação no ângulo de 80° com n=1') plt.xlabel('Ângulo do Encoder') plt.ylabel('Ângulo do Aplicativo') plt.xlim(80, 80.2) plt.ylim(79.50, 82.1) plt.grid(True) plt.savefig('PDF/AmpliacaoCelularMatheusN1.pdf', format='pdf', bbox_inches = 'tight') plt.savefig('EPS/AmpliacaoCelularMatheusN1.eps', format='eps', bbox_inches = 'tight') #EXIBINDO plt.show() <jupyter_output><empty_output><jupyter_text>### n=9<jupyter_code>#PLOTANDO A IDEAL plt.scatter(encoder9, encoder9, s=100, alpha=0.7, label="IDEAL", color="crimson") #PLOTANDO A COMPARAÇÃO plt.scatter(encoder9, app9, s=70, label="OBTIDO", color="darkblue") #OUTROS PARÂMETROS DO GRÁFICO plt.legend() plt.title('Comparação com n=9') plt.xlabel('Ângulo do Encoder') plt.ylabel('Ângulo do Aplicativo') plt.xlim(0, 90) plt.ylim(0, 90) plt.xticks(array([0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]), ['0°', '10°', '20°', '30°', '40°', '50°', '60°', '70°', '80°', '90°']) plt.yticks(array([0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]), ['0°', '10°', '20°', '30°', '40°', '50°', '60°', '70°', '80°', '90°']) plt.grid(True) plt.savefig('PDF/celularMatheusN9.pdf', format='pdf', bbox_inches = 'tight') plt.savefig('EPS/celularMatheusN9.eps', format='eps', bbox_inches = 'tight') #EXIBINDO plt.show() #PLOTANDO A IDEAL plt.errorbar(encoder9, encoder9, yerr=ictzEnc9, alpha=0.7, label="IDEAL", color="crimson", fmt='o') #PLOTANDO A COMPARAÇÃO plt.errorbar(encoder9, app9, yerr=ictzApp9, label="OBTIDO", color="darkblue", fmt='o') #OUTROS PARÂMETROS DO GRÁFICO plt.legend() plt.title('Ampliação no ângulo de 80° com n=9') plt.xlabel('Ângulo do Encoder') plt.ylabel('Ângulo do Aplicativo') plt.xlim(40, 40.55) plt.ylim(38.5, 40.75) plt.grid(True) plt.savefig('PDF/AmpliacaoCelularMatheusN9.pdf', format='pdf', bbox_inches = 'tight') plt.savefig('EPS/AmpliacaoCelularMatheusN9.eps', format='eps', bbox_inches = 'tight') #EXIBINDO plt.show() <jupyter_output><empty_output><jupyter_text>## ERROS NORMALIZADOS#### n = 1<jupyter_code># GERANDO O ARRAY DE ERROS NORMALIZADOS PARA N = 1 erroNormal1 = [] pontilhada = [] cor = [] for i in range(0, len(app1)): dividendo = absolute(app1[i] - encoder1[i]) divisor = sqrt((2*ictzApp1[i])**2 + (2*ictzEnc1[i])**2) erroNormalizadoAtual = dividendo / divisor erroNormal1.append(erroNormalizadoAtual) pontilhada.append(1) cor.append("green") erroNormal1 = array(erroNormal1) pontilhada = array(pontilhada) #PLOTANDO OS ERROS plt.scatter(encoder1, erroNormal1, s=150, alpha=0.8, label="ERRO NORMALIZADO", color="darkblue") #PLOTANDO A PONTILHADA plt.plot(encoder1, pontilhada, '--', color="black") #OUTROS PARÂMETROS DO GRÁFICO plt.legend() plt.title('Erros Normalizados com n=1') plt.xlabel('Ângulo') plt.ylabel('Erro Normalizado') plt.xlim(0, 90) plt.ylim(0, 5) plt.xticks(array([0, 10, 20, 30, 40, 50, 60, 70, 80, 90]), ['0°', '10°', '20°', '30°', '40°', '50°', '60°', '70°', '80°', '90°']) plt.grid(True) plt.savefig('PDF/ErroNormalizadoN1.pdf', format='pdf', bbox_inches = 'tight') plt.savefig('EPS/ErroNormalizadoN1.eps', format='eps', bbox_inches = 'tight') plt.plot()<jupyter_output><empty_output><jupyter_text>#### n = 9<jupyter_code># GERANDO O ARRAY DE ERROS NORMALIZADOS PARA N = 9 erroNormal9 = [] pontilhada = [] for i in range(0, len(app9)): dividendo = absolute(app9[i] - encoder9[i]) divisor = sqrt((2*ictzApp9[i])**2 + (2*ictzEnc9[i])**2) erroNormalizadoAtual = dividendo / divisor erroNormal9.append(erroNormalizadoAtual) pontilhada.append(1) erroNormal9 = array(erroNormal9) pontilhada = array(pontilhada) #PLOTANDO OS ERROS plt.scatter(encoder9, erroNormal9, s=150, alpha=0.8, label="ERRO NORMALIZADO", color="darkblue") #PLOTANDO A PONTILHADA plt.plot(encoder9, pontilhada, '--', color="black") #OUTROS PARÂMETROS DO GRÁFICO plt.legend() plt.title('Erros Normalizados com n=9') plt.xlabel('Ângulo') plt.ylabel('Erro Normalizado') plt.xlim(0, 90) plt.ylim(0, 5) plt.xticks(array([0, 10, 20, 30, 40, 50, 60, 70, 80, 90]), ['0°', '10°', '20°', '30°', '40°', '50°', '60°', '70°', '80°', '90°']) plt.grid(True) plt.savefig('PDF/ErroNormalizadoN9.pdf', format='pdf', bbox_inches = 'tight') plt.savefig('EPS/ErroNormalizadoN9.eps', format='eps', bbox_inches = 'tight') plt.plot()<jupyter_output><empty_output>
non_permissive
/Aquisicoes/Primeira Tentativa/30.04.2019/Primeira_Tentativa.ipynb
Mathcd/MEE-sensorsProject
8
<jupyter_start><jupyter_text>Preprocesing<jupyter_code>netflix_titles_df = pd.read_csv('netflix_titles.csv') netflix_titles_df.head() netflix_titles_df.drop(netflix_titles_df.columns[[0,1,5,6,7,9]], axis=1, inplace=True) netflix_titles_df.count() null_rows = len(netflix_titles_df[netflix_titles_df.isna().any(axis=1)]) print(f'Rows with NaNs: {null_rows} ({(null_rows/netflix_titles_df.shape[0])*100:.0f}%)') netflix_titles_df.fillna('', inplace=True) netflix_titles_df.head() netflix_titles_df[['director','cast']] = netflix_titles_df[['director','cast']].applymap(lambda x: ' '.join(x.replace(' ', '').split(',')[:3])) netflix_titles_df.head() netflix_titles_df['title_dup'] = netflix_titles_df['title'] titles_corpus = netflix_titles_df.apply(' '.join, axis=1) titles_corpus.head tfidf_vectorizer_params = TfidfVectorizer(lowercase=True, stop_words='english', ngram_range=(1, 3), max_df = .5)<jupyter_output><empty_output><jupyter_text>Text Vectorization <jupyter_code>tfidf_vectorizer = tfidf_vectorizer_params.fit_transform(titles_corpus) pd.DataFrame(tfidf_vectorizer.toarray(), columns=tfidf_vectorizer_params.get_feature_names_out()) pickle.dump(tfidf_vectorizer, open('tfidf_vectorizer.pickle', 'wb')) vects_cos_sim = cosine_similarity(tfidf_vectorizer, tfidf_vectorizer) pd.DataFrame(data=vects_cos_sim, index=netflix_titles_df['title'], columns=netflix_titles_df['title']).head()<jupyter_output><empty_output>
no_license
/Netflix_Shows_Recommendation_API/preprocessing.ipynb
AnantShankhdhar/Recommender-Systems
2
<jupyter_start><jupyter_text># Pretrained BERT models<jupyter_code>import sys package_dir = "../input/ppbert/pytorch-pretrained-bert/pytorch-pretrained-BERT" sys.path.append(package_dir) from __future__ import absolute_import from __future__ import division from __future__ import print_function import torch.utils.data import numpy as np import pandas as pd from tqdm import tqdm import os import warnings from pytorch_pretrained_bert import BertTokenizer, BertForSequenceClassification, BertAdam from pytorch_pretrained_bert import BertConfig import gc warnings.filterwarnings(action='once') device = torch.device('cuda') def convert_lines(example, max_seq_length,tokenizer): max_seq_length -=2 all_tokens = [] longer = 0 for text in tqdm(example): tokens_a = tokenizer.tokenize(text) if len(tokens_a)>max_seq_length: tokens_a = tokens_a[:max_seq_length] longer += 1 one_token = tokenizer.convert_tokens_to_ids(["[CLS]"]+tokens_a+["[SEP]"])+[0] * (max_seq_length - len(tokens_a)) all_tokens.append(one_token) return np.array(all_tokens) MAX_SEQUENCE_LENGTH = 220 SEED = 1234 BATCH_SIZE = 32 BERT_MODEL_PATH = '../input/bert-pretrained-models/uncased_l-12_h-768_a-12/uncased_L-12_H-768_A-12/' LARGE_BERT_MODEL_PATH = '../input/bert-pretrained-models/uncased_l-24_h-1024_a-16/uncased_L-24_H-1024_A-16/' np.random.seed(SEED) torch.manual_seed(SEED) torch.cuda.manual_seed(SEED) torch.backends.cudnn.deterministic = True # Pretrained BERT models - Google's pretrained BERT model BERT_SMALL_PATH = '../input/bert-pretrained-models/uncased_l-12_h-768_a-12/uncased_L-12_H-768_A-12/' BERT_LARGE_PATH = '../input/bert-pretrained-models/uncased_l-24_h-1024_a-16/uncased_L-24_H-1024_A-16/' # JIGSAW fine-tuned BERT models JIGSAW_BERT_SMALL_MODEL_PATH = '../input/bert-inference/bert/bert_pytorch.bin' JIGSAW_BERT_LARGE_MODEL_PATH = '../input/jigsawpretrainedbertmodels/jigsaw-bert-large-uncased-len-220-fp16/epoch-1/pytorch_model.bin' JIGSAW_BERT_SMALL_JSON_PATH = '../input/bert-inference/bert/bert_config.json' JIGSAW_BERT_LARGE_JSON_PATH = '../input/jigsawpretrainedbertmodels/jigsaw-bert-large-uncased-len-220-fp16/epoch-1/config.json' NUM_BERT_MODELS = 2 INFER_BATCH_SIZE = 64 train_df = pd.read_csv('../input/jigsaw-unintended-bias-in-toxicity-classification/train.csv') test_df = pd.read_csv('../input/jigsaw-unintended-bias-in-toxicity-classification/test.csv') test_preds = np.zeros((test_df.shape[0],NUM_BERT_MODELS)) np.random.seed(SEED) torch.manual_seed(SEED) torch.cuda.manual_seed(SEED) torch.backends.cudnn.deterministic = True print("Predicting BERT large model......") # Prepare data bert_config = BertConfig(JIGSAW_BERT_LARGE_JSON_PATH) tokenizer = BertTokenizer.from_pretrained(BERT_LARGE_PATH, cache_dir=None,do_lower_case=True) X_test = convert_lines(test_df["comment_text"].fillna("DUMMY_VALUE"), MAX_SEQUENCE_LENGTH, tokenizer) test = torch.utils.data.TensorDataset(torch.tensor(X_test, dtype=torch.long)) # Load fine-tuned BERT model gc.collect() model = BertForSequenceClassification(bert_config, num_labels=1) model.load_state_dict(torch.load(JIGSAW_BERT_LARGE_MODEL_PATH)) model.to(device) for param in model.parameters(): param.requires_grad = False model.eval() # Predicting model_preds = np.zeros((len(X_test))) test_loader = torch.utils.data.DataLoader(test, batch_size=INFER_BATCH_SIZE, shuffle=False) tk0 = tqdm(test_loader) for i, (x_batch,) in enumerate(tk0): pred = model(x_batch.to(device), attention_mask=(x_batch > 0).to(device), labels=None) model_preds[i * INFER_BATCH_SIZE:(i + 1) * INFER_BATCH_SIZE] = pred[:, 0].detach().cpu().squeeze().numpy() test_preds[:,0] = torch.sigmoid(torch.tensor(model_preds)).numpy().ravel() del model gc.collect() print("Predicting BERT small model......") bert_config = BertConfig(JIGSAW_BERT_SMALL_JSON_PATH) tokenizer = BertTokenizer.from_pretrained(BERT_SMALL_PATH, cache_dir=None,do_lower_case=True) X_test = convert_lines(test_df["comment_text"].fillna("DUMMY_VALUE"), MAX_SEQUENCE_LENGTH, tokenizer) test = torch.utils.data.TensorDataset(torch.tensor(X_test, dtype=torch.long)) # # # Load fine-tuned BERT model model = BertForSequenceClassification(bert_config, num_labels=1) model.load_state_dict(torch.load(JIGSAW_BERT_SMALL_MODEL_PATH)) model.to(device) for param in model.parameters(): param.requires_grad = False model.eval() # Predicting model_preds = np.zeros((len(X_test))) test_loader = torch.utils.data.DataLoader(test, batch_size=INFER_BATCH_SIZE, shuffle=False) tk0 = tqdm(test_loader) for i, (x_batch,) in enumerate(tk0): pred = model(x_batch.to(device), attention_mask=(x_batch > 0).to(device), labels=None) model_preds[i * INFER_BATCH_SIZE:(i + 1) * INFER_BATCH_SIZE] = pred[:, 0].detach().cpu().squeeze().numpy() test_preds[:,1] = torch.sigmoid(torch.tensor(model_preds)).numpy().ravel() del model gc.collect() # Sub-model prediction bert_submission = pd.DataFrame.from_dict({ 'id': test_df['id'], 'prediction': test_preds.mean(axis=1)}) bert_submission.to_csv('bert_submission.csv', index=False) <jupyter_output><empty_output><jupyter_text>**Credits** This notebook was mainly inspired by the following awesome kernel scripts: https://www.kaggle.com/gpreda/jigsaw-fast-compact-solution https://www.kaggle.com/christofhenkel/how-to-preprocessing-for-glove-part2-usage https://www.kaggle.com/shubham505/apply-by-simple-bilstm # Preparations ## Datasets You will need to add the following Kaggle dataset for pickled pretrained embeddings https://www.kaggle.com/chriscc/pickled-word-embedding ## Import packages<jupyter_code># This Python 3 environment comes with many helpful analytics libraries installed # It is defined by the kaggle/python docker image: https://github.com/kaggle/docker-python # For example, here's several helpful packages to load in import numpy as np # linear algebra import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv) import gc import re # Input data files are available in the "../input/" directory. # For example, running this (by clicking run or pressing Shift+Enter) will list the files in the input directory import os print(os.listdir("../input")) from keras.preprocessing import text, sequence from keras import backend as K from keras.models import Model from keras.layers import Input, Dense, Embedding, SpatialDropout1D, add, concatenate from keras.layers import CuDNNLSTM, Bidirectional, GlobalMaxPooling1D, GlobalAveragePooling1D from keras.preprocessing import text, sequence from keras.callbacks import LearningRateScheduler from keras.engine.topology import Layer from keras import initializers, regularizers, constraints, optimizers, layers from tqdm._tqdm_notebook import tqdm_notebook as tqdm import pickle tqdm.pandas()<jupyter_output><empty_output><jupyter_text>## Configurations<jupyter_code>EMBEDDING_PATHS = ['../input/pickled-word-embedding/crawl-300d-2M.pkl', '../input/pickled-word-embedding/glove.840B.300d.pkl'] NUM_MODELS = 2 # The number of classifiers we want to train BATCH_SIZE = 512 # can be tuned LSTM_UNITS = 128 # can be tuned DENSE_HIDDEN_UNITS = 4*LSTM_UNITS # can betuned EPOCHS = 4 # The number of epoches we want to train for each classifier MAX_LEN = 220 # can ben tuned # in bert + lstm blend kernel was identity_columns = [ # 'male', 'female', 'homosexual_gay_or_lesbian', 'christian', 'jewish', # 'muslim', 'black', 'white', 'psychiatric_or_mental_illness'] IDENTITY_COLUMNS = [ 'transgender', 'female', 'homosexual_gay_or_lesbian', 'muslim', 'hindu', 'white', 'black', 'psychiatric_or_mental_illness', 'jewish' ] AUX_COLUMNS = ['target', 'severe_toxicity','obscene','identity_attack','insult','threat'] TEXT_COLUMN = 'comment_text' TARGET_COLUMN = 'target'<jupyter_output><empty_output><jupyter_text>## Utils<jupyter_code>#----------------------------------- Preprocessing-------------------------------------# SYMBOLS_TO_ISOLATE = '.,?!-;*"…:—()%#$&_/@\・ω+=”“[]^–>\\°<~•≠™ˈʊɒ∞§{}·τα❤☺ɡ|¢→̶`❥━┣┫┗O►★©―ɪ✔®\x96\x92●£♥➤´¹☕≈÷♡◐║▬′ɔː€۩۞†μ✒➥═☆ˌ◄½ʻπδηλσερνʃ✬SUPERIT☻±♍µº¾✓◾؟.⬅℅»Вав❣⋅¿¬♫CMβ█▓▒░⇒⭐›¡₂₃❧▰▔◞▀▂▃▄▅▆▇↙γ̄″☹➡«φ⅓„✋:¥̲̅́∙‛◇✏▷❓❗¶˚˙)сиʿ✨。ɑ\x80◕!%¯−flfi₁²ʌ¼⁴⁄₄⌠♭✘╪▶☭✭♪☔☠♂☃☎✈✌✰❆☙○‣⚓年∎ℒ▪▙☏⅛casǀ℮¸w‚∼‖ℳ❄←☼⋆ʒ⊂、⅔¨͡๏⚾⚽Φ×θ₩?(℃⏩☮⚠月✊❌⭕▸■⇌☐☑⚡☄ǫ╭∩╮,例>ʕɐ̣Δ₀✞┈╱╲▏▕┃╰▊▋╯┳┊≥☒↑☝ɹ✅☛♩☞AJB◔◡↓♀⬆̱ℏ\x91⠀ˤ╚↺⇤∏✾◦♬³の|/∵∴√Ω¤☜▲↳▫‿⬇✧ovm-208'‰≤∕ˆ⚜☁' SYMBOLS_TO_REMOVE = '\n🍕\r🐵\xa0\ue014\t\uf818\uf04a\xad😢🐶️\uf0e0😜😎👊\u200b\u200e😁عدويهصقأناخلىبمغر😍💖💵Е👎😀😂\u202a\u202c🔥😄🏻💥ᴍʏʀᴇɴᴅᴏᴀᴋʜᴜʟᴛᴄᴘʙғᴊᴡɢ😋👏שלוםבי😱‼\x81エンジ故障\u2009🚌ᴵ͞🌟😊😳😧🙀😐😕\u200f👍😮😃😘אעכח💩💯⛽🚄🏼ஜ😖ᴠ🚲‐😟😈💪🙏🎯🌹😇💔😡\x7f👌ἐὶήιὲκἀίῃἴξ🙄H😠\ufeff\u2028😉😤⛺🙂\u3000تحكسة👮💙فزط😏🍾🎉😞\u2008🏾😅😭👻😥😔😓🏽🎆🍻🍽🎶🌺🤔😪\x08‑🐰🐇🐱🙆😨🙃💕𝘊𝘦𝘳𝘢𝘵𝘰𝘤𝘺𝘴𝘪𝘧𝘮𝘣💗💚地獄谷улкнПоАН🐾🐕😆ה🔗🚽歌舞伎🙈😴🏿🤗🇺🇸мυтѕ⤵🏆🎃😩\u200a🌠🐟💫💰💎эпрд\x95🖐🙅⛲🍰🤐👆🙌\u2002💛🙁👀🙊🙉\u2004ˢᵒʳʸᴼᴷᴺʷᵗʰᵉᵘ\x13🚬🤓\ue602😵άοόςέὸתמדףנרךצט😒͝🆕👅👥👄🔄🔤👉👤👶👲🔛🎓\uf0b7\uf04c\x9f\x10成都😣⏺😌🤑🌏😯ех😲Ἰᾶὁ💞🚓🔔📚🏀👐\u202d💤🍇\ue613小土豆🏡❔⁉\u202f👠》कर्मा🇹🇼🌸蔡英文🌞🎲レクサス😛外国人关系Сб💋💀🎄💜🤢َِьыгя不是\x9c\x9d🗑\u2005💃📣👿༼つ༽😰ḷЗз▱ц🤣卖温哥华议会下降你失去所有的钱加拿大坏税骗子🐝ツ🎅\x85🍺آإشء🎵🌎͟ἔ油别克🤡🤥😬🤧й\u2003🚀🤴ʲшчИОРФДЯМюж😝🖑ὐύύ特殊作戦群щ💨圆明园קℐ🏈😺🌍⏏ệ🍔🐮🍁🍆🍑🌮🌯🤦\u200d𝓒𝓲𝓿𝓵안영하세요ЖљКћ🍀😫🤤ῦ我出生在了可以说普通话汉语好极🎼🕺🍸🥂🗽🎇🎊🆘🤠👩🖒🚪天一家⚲\u2006⚭⚆⬭⬯⏖新✀╌🇫🇷🇩🇪🇮🇬🇧😷🇨🇦ХШ🌐\x1f杀鸡给猴看ʁ𝗪𝗵𝗲𝗻𝘆𝗼𝘂𝗿𝗮𝗹𝗶𝘇𝗯𝘁𝗰𝘀𝘅𝗽𝘄𝗱📺ϖ\u2000үսᴦᎥһͺ\u2007հ\u2001ɩye൦lƽh𝐓𝐡𝐞𝐫𝐮𝐝𝐚𝐃𝐜𝐩𝐭𝐢𝐨𝐧Ƅᴨןᑯ໐ΤᏧ௦Іᴑ܁𝐬𝐰𝐲𝐛𝐦𝐯𝐑𝐙𝐣𝐇𝐂𝐘𝟎ԜТᗞ౦〔Ꭻ𝐳𝐔𝐱𝟔𝟓𝐅🐋ffi💘💓ё𝘥𝘯𝘶💐🌋🌄🌅𝙬𝙖𝙨𝙤𝙣𝙡𝙮𝙘𝙠𝙚𝙙𝙜𝙧𝙥𝙩𝙪𝙗𝙞𝙝𝙛👺🐷ℋ𝐀𝐥𝐪🚶𝙢Ἱ🤘ͦ💸ج패티W𝙇ᵻ👂👃ɜ🎫\uf0a7БУі🚢🚂ગુજરાતીῆ🏃𝓬𝓻𝓴𝓮𝓽𝓼☘﴾̯﴿₽\ue807𝑻𝒆𝒍𝒕𝒉𝒓𝒖𝒂𝒏𝒅𝒔𝒎𝒗𝒊👽😙\u200cЛ‒🎾👹⎌🏒⛸公寓养宠物吗🏄🐀🚑🤷操美𝒑𝒚𝒐𝑴🤙🐒欢迎来到阿拉斯ספ𝙫🐈𝒌𝙊𝙭𝙆𝙋𝙍𝘼𝙅ﷻ🦄巨收赢得白鬼愤怒要买额ẽ🚗🐳𝟏𝐟𝟖𝟑𝟕𝒄𝟗𝐠𝙄𝙃👇锟斤拷𝗢𝟳𝟱𝟬⦁マルハニチロ株式社⛷한국어ㄸㅓ니͜ʖ𝘿𝙔₵𝒩ℯ𝒾𝓁𝒶𝓉𝓇𝓊𝓃𝓈𝓅ℴ𝒻𝒽𝓀𝓌𝒸𝓎𝙏ζ𝙟𝘃𝗺𝟮𝟭𝟯𝟲👋🦊多伦🐽🎻🎹⛓🏹🍷🦆为和中友谊祝贺与其想象对法如直接问用自己猜本传教士没积唯认识基督徒曾经让相信耶稣复活死怪他但当们聊些政治题时候战胜因圣把全堂结婚孩恐惧且栗谓这样还♾🎸🤕🤒⛑🎁批判检讨🏝🦁🙋😶쥐스탱트뤼도석유가격인상이경제황을렵게만들지않록잘관리해야합다캐나에서대마초와화약금의품런성분갈때는반드시허된사용🔫👁凸ὰ💲🗯𝙈Ἄ𝒇𝒈𝒘𝒃𝑬𝑶𝕾𝖙𝖗𝖆𝖎𝖌𝖍𝖕𝖊𝖔𝖑𝖉𝖓𝖐𝖜𝖞𝖚𝖇𝕿𝖘𝖄𝖛𝖒𝖋𝖂𝕴𝖟𝖈𝕸👑🚿💡知彼百\uf005𝙀𝒛𝑲𝑳𝑾𝒋𝟒😦𝙒𝘾𝘽🏐𝘩𝘨ὼṑ𝑱𝑹𝑫𝑵𝑪🇰🇵👾ᓇᒧᔭᐃᐧᐦᑳᐨᓃᓂᑲᐸᑭᑎᓀᐣ🐄🎈🔨🐎🤞🐸💟🎰🌝🛳点击查版🍭𝑥𝑦𝑧NG👣\uf020っ🏉ф💭🎥Ξ🐴👨🤳🦍\x0b🍩𝑯𝒒😗𝟐🏂👳🍗🕉🐲چی𝑮𝗕𝗴🍒ꜥⲣⲏ🐑⏰鉄リ事件ї💊「」\uf203\uf09a\uf222\ue608\uf202\uf099\uf469\ue607\uf410\ue600燻製シ虚偽屁理屈Г𝑩𝑰𝒀𝑺🌤𝗳𝗜𝗙𝗦𝗧🍊ὺἈἡχῖΛ⤏🇳𝒙ψՁմեռայինրւդձ冬至ὀ𝒁🔹🤚🍎𝑷🐂💅𝘬𝘱𝘸𝘷𝘐𝘭𝘓𝘖𝘹𝘲𝘫کΒώ💢ΜΟΝΑΕ🇱♲𝝈↴💒⊘Ȼ🚴🖕🖤🥘📍👈➕🚫🎨🌑🐻𝐎𝐍𝐊𝑭🤖🎎😼🕷grntidufbk𝟰🇴🇭🇻🇲𝗞𝗭𝗘𝗤👼📉🍟🍦🌈🔭《🐊🐍\uf10aლڡ🐦\U0001f92f\U0001f92a🐡💳ἱ🙇𝗸𝗟𝗠𝗷🥜さようなら🔼' ISOLATE_DICT = {ord(c):f' {c} ' for c in SYMBOLS_TO_ISOLATE} REMOVE_DICT = {ord(c):f'' for c in SYMBOLS_TO_REMOVE} CHARS_TO_REMOVE = '!"#$%&()*+,-./:;<=>?@[\\]^_`{|}~\t\n“”’\'∞θ÷α•à−β∅³π‘₹´°£€\×™√²—' CONTRACTION_MAPPING = {"ain't": "is not", "aren't": "are not","can't": "cannot", "'cause": "because", "could've": "could have", "couldn't": "could not", "didn't": "did not", "doesn't": "does not", "don't": "do not", "hadn't": "had not", "hasn't": "has not", "haven't": "have not", "he'd": "he would","he'll": "he will", "he's": "he is", "how'd": "how did", "how'd'y": "how do you", "how'll": "how will", "how's": "how is", "I'd": "I would", "I'd've": "I would have", "I'll": "I will", "I'll've": "I will have","I'm": "I am", "I've": "I have", "i'd": "i would", "i'd've": "i would have", "i'll": "i will", "i'll've": "i will have","i'm": "i am", "i've": "i have", "isn't": "is not", "it'd": "it would", "it'd've": "it would have", "it'll": "it will", "it'll've": "it will have","it's": "it is", "let's": "let us", "ma'am": "madam", "mayn't": "may not", "might've": "might have","mightn't": "might not","mightn't've": "might not have", "must've": "must have", "mustn't": "must not", "mustn't've": "must not have", "needn't": "need not", "needn't've": "need not have","o'clock": "of the clock", "oughtn't": "ought not", "oughtn't've": "ought not have", "shan't": "shall not", "sha'n't": "shall not", "shan't've": "shall not have", "she'd": "she would", "she'd've": "she would have", "she'll": "she will", "she'll've": "she will have", "she's": "she is", "should've": "should have", "shouldn't": "should not", "shouldn't've": "should not have", "so've": "so have","so's": "so as", "this's": "this is","that'd": "that would", "that'd've": "that would have", "that's": "that is", "there'd": "there would", "there'd've": "there would have", "there's": "there is", "here's": "here is","they'd": "they would", "they'd've": "they would have", "they'll": "they will", "they'll've": "they will have", "they're": "they are", "they've": "they have", "to've": "to have", "wasn't": "was not", "we'd": "we would", "we'd've": "we would have", "we'll": "we will", "we'll've": "we will have", "we're": "we are", "we've": "we have", "weren't": "were not", "what'll": "what will", "what'll've": "what will have", "what're": "what are", "what's": "what is", "what've": "what have", "when's": "when is", "when've": "when have", "where'd": "where did", "where's": "where is", "where've": "where have", "who'll": "who will", "who'll've": "who will have", "who's": "who is", "who've": "who have", "why's": "why is", "why've": "why have", "will've": "will have", "won't": "will not", "won't've": "will not have", "would've": "would have", "wouldn't": "would not", "wouldn't've": "would not have", "y'all": "you all", "y'all'd": "you all would","y'all'd've": "you all would have","y'all're": "you all are","y'all've": "you all have","you'd": "you would", "you'd've": "you would have", "you'll": "you will", "you'll've": "you will have", "you're": "you are", "you've": "you have" } def handle_punctuation(text): text = text.translate(REMOVE_DICT) text = text.translate(ISOLATE_DICT) return text def clean_contractions(text, mapping=CONTRACTION_MAPPING): ''' Expand contractions ''' specials = ["’", "‘", "´", "`"] for s in specials: text = text.replace(s, "'") text = ' '.join([mapping[t] if t in mapping else t for t in text.split(" ")]) return text def preprocess(x): x = handle_punctuation(x) # x = clean_contractions(x) return x #----------------------------------- Embedding -------------------------------------# def get_coefs(word, *arr): """ Get word, word_embedding from a pretrained embedding file """ return word, np.asarray(arr,dtype='float32') def load_embeddings(path): if path.split('.')[-1] in ['txt','vec']: # for original pretrained embedding files (extension .text, .vec) with open(path,'rb') as f: return dict(get_coefs(*line.strip().split(' ')) for line in f) if path.split('.')[-1] =='pkl': # for pickled pretrained embedding files (extention pkl). Loading pickeled embeddings is faster than texts with open(path,'rb') as f: return pickle.load(f) def build_matrix(word_index, path): """ Here we take each word we've tokenized in our text corpus for each word we look up in the pre-trained embedding. Each row in this matrix is a corpus word's embedding. """ embedding_index = load_embeddings(path) embedding_matrix = np.zeros((len(word_index)+1, 300)) for word, i in word_index.items(): try: embedding_matrix[i] = embedding_index[word] except KeyError: pass return embedding_matrix<jupyter_output><empty_output><jupyter_text>## Define LSTM model<jupyter_code>class Attention(Layer): def __init__(self, step_dim, W_regularizer=None, b_regularizer=None, W_constraint=None, b_constraint=None, bias=True, **kwargs): self.supports_masking = True self.init = initializers.get('glorot_uniform') self.W_regularizer = regularizers.get(W_regularizer) self.b_regularizer = regularizers.get(b_regularizer) self.W_constraint = constraints.get(W_constraint) self.b_constraint = constraints.get(b_constraint) self.bias = bias self.step_dim = step_dim self.features_dim = 0 super(Attention, self).__init__(**kwargs) def build(self, input_shape): assert len(input_shape) == 3 self.W = self.add_weight((input_shape[-1],), initializer=self.init, name='{}_W'.format(self.name), regularizer=self.W_regularizer, constraint=self.W_constraint) self.features_dim = input_shape[-1] if self.bias: self.b = self.add_weight((input_shape[1],), initializer='zero', name='{}_b'.format(self.name), regularizer=self.b_regularizer, constraint=self.b_constraint) else: self.b = None self.built = True def compute_mask(self, input, input_mask=None): return None def call(self, x, mask=None): features_dim = self.features_dim step_dim = self.step_dim eij = K.reshape(K.dot(K.reshape(x, (-1, features_dim)), K.reshape(self.W, (features_dim, 1))), (-1, step_dim)) if self.bias: eij += self.b eij = K.tanh(eij) a = K.exp(eij) if mask is not None: a *= K.cast(mask, K.floatx()) a /= K.cast(K.sum(a, axis=1, keepdims=True) + K.epsilon(), K.floatx()) a = K.expand_dims(a) weighted_input = x * a return K.sum(weighted_input, axis=1) def compute_output_shape(self, input_shape): return input_shape[0], self.features_dim def build_model(embedding_matrix, num_aux_targets):#, loss_weight): """ embedding layer droput layer 2 * bidirectional LSTM layers 2 * pooling layers 2 dense layers 1 softmax layer """ words = Input(shape=(MAX_LEN,)) #Embedding layer takes variable size input x = Embedding(*embedding_matrix.shape, weights = [embedding_matrix], trainable=False)(words) x = SpatialDropout1D(0.2)(x) x = Bidirectional(CuDNNLSTM(LSTM_UNITS, return_sequences=True))(x) x = Bidirectional(CuDNNLSTM(LSTM_UNITS, return_sequences=True))(x) #att = Attention(MAX_LEN)(x) hidden = concatenate([ GlobalMaxPooling1D()(x), GlobalAveragePooling1D()(x) ]) hidden = add([hidden, Dense(DENSE_HIDDEN_UNITS, activation='relu')(hidden)]) hidden = add([hidden, Dense(DENSE_HIDDEN_UNITS, activation='relu')(hidden)]) result = Dense(1, activation='sigmoid')(hidden) aux_result =Dense(num_aux_targets, activation='sigmoid')(hidden) model = Model(inputs =words, outputs =[result, aux_result]) model.compile(loss='binary_crossentropy', optimizer='adam') return model<jupyter_output><empty_output><jupyter_text># Data preprocessing<jupyter_code># Preprocess comment texts train_df['comment_text'] = train_df['comment_text'].progress_apply(lambda x:preprocess(x)) test_df['comment_text'] = test_df['comment_text'].progress_apply(lambda x:preprocess(x)) gc.collect() x_train = train_df[TEXT_COLUMN].astype(str) y_train = train_df[TARGET_COLUMN].values y_aux_train = train_df[AUX_COLUMNS].values x_test = test_df[TEXT_COLUMN].astype(str) # Convert target probability to 1 or 0 so they can be used for classification for column in IDENTITY_COLUMNS + [TARGET_COLUMN]: train_df[column] = np.where(train_df[column] >=0.5, True, False)<jupyter_output><empty_output><jupyter_text># Tokenize comment texts<jupyter_code># Return a Keras tokenizer class tokenizer = text.Tokenizer(filters = CHARS_TO_REMOVE) tokenizer.fit_on_texts(list(x_train)+ list(x_test)) # Turn text to sequences of tokens x_train = tokenizer.texts_to_sequences(x_train) x_test = tokenizer.texts_to_sequences(x_test) #Pad sequences to the same length x_train = sequence.pad_sequences(x_train,maxlen=MAX_LEN) x_test= sequence.pad_sequences(x_test, maxlen=MAX_LEN) # Initialize weights sample_weights = np.ones(len(x_train), dtype=np.float32) # Add all the values of the identities along rows sample_weights += train_df[IDENTITY_COLUMNS].sum(axis=1) #Add all values of targets*~identity sample_weights += train_df[TARGET_COLUMN]*(~train_df[IDENTITY_COLUMNS]).sum(axis=1) #Add all values ~targets*identity sample_weights += (~train_df[TARGET_COLUMN])*train_df[IDENTITY_COLUMNS].sum(axis=1) #Normalize them sample_weights/=sample_weights.mean()<jupyter_output><empty_output><jupyter_text>## Create embedding matrix<jupyter_code>embedding_matrix = np.concatenate([build_matrix(tokenizer.word_index,f) for f in EMBEDDING_PATHS], axis =-1) print("Embedding matrix shape:", embedding_matrix.shape) del train_df, tokenizer gc.collect()<jupyter_output><empty_output><jupyter_text># Model training * 2 models will be trained (NUM_MODELS=2) * Make predictions at the end of each epoch * Weighted averaging epoch predictions * Weights = 2 ** epoch<jupyter_code>checkpoint_predictions = [] weights = [] NUM_MODELS = 1 for model_idx in range(NUM_MODELS): #Passes embedding matrix and aux outputs shape model = build_model(embedding_matrix, y_aux_train.shape[-1]) #1/sample_weights.mean()) for global_epoch in range(EPOCHS): model.fit( x_train, [y_train, y_aux_train], batch_size=BATCH_SIZE, epochs=1, verbose=1, sample_weight=[sample_weights.values, np.ones_like(sample_weights)], callbacks = [ LearningRateScheduler(lambda _: 1e-3*(0.55**global_epoch)) # Decayed learning rate ] ) # model.save_weights("model_%d_%d.h5" % (model_idx, global_epoch)) # Save model weights checkpoint_predictions.append(model.predict(x_test, batch_size=2048)[0].flatten()) weights.append(2 ** global_epoch) del model # If a model didn't get deleted Keras will continue training it even though build_model() was used to initialize a model gc.collect() # It's a good practice to use gc.collect() once the training is done to free up RAM print (weights) predictions = np.average(checkpoint_predictions, weights=weights, axis=0) lstm_submission = pd.DataFrame.from_dict({ 'id': test_df.id, 'prediction': predictions }) lstm_submission.to_csv('submission.csv', index=False) submission = pd.DataFrame.from_dict({ 'id': test_df['id'], 'prediction': lstm_submission['prediction'].rank(pct=True)*0.4 + bert_submission['prediction'].rank(pct=True)*0.6}) submission.to_csv('submission.csv', index=False) <jupyter_output><empty_output>
no_license
/jigsaw-starter-blend.ipynb
OmerBgu/jigsaw-unintended-bias-in-toxicity-classification
9
<jupyter_start><jupyter_text>Table of Contents 1&nbsp;&nbsp;Reuso de Código1.1&nbsp;&nbsp;Pacotes e Módulos1.2&nbsp;&nbsp;Um pouco sobre pacotes1.3&nbsp;&nbsp;Localizando Módulos2&nbsp;&nbsp;Módulos personalizados2.1&nbsp;&nbsp;O módulo vector.py2.1.1&nbsp;&nbsp;Vector Add2.1.2&nbsp;&nbsp;Vector Subtraction2.1.3&nbsp;&nbsp;Vector Multiplication2.1.4&nbsp;&nbsp;Vector Division2.1.5&nbsp;&nbsp;Vector Magnitude2.1.6&nbsp;&nbsp;Vector Normalization2.2&nbsp;&nbsp;A Classe Vector2.3&nbsp;&nbsp;Definições de Métodos2.3.1&nbsp;&nbsp;O Método __init__:2.3.2&nbsp;&nbsp;O Método __str__:2.3.3&nbsp;&nbsp;Outros métodos especiais:2.3.4&nbsp;&nbsp;Métodos normais de uma classe2.4&nbsp;&nbsp;Exercícios# Reuso de CódigoSe tivéssemos que escrever todos os programas a partir do zero, não poderíamos fazer muito. Parte da diversão da programação é usar algo que outra pessoa escreveu para resolver um problema mais rapidamente. Outro aspecto divertido da programação é escrever o código que outros possam reutilizar em seus programas. Considerando que as funções nos permitem "compor" partes de código para que possam ser reutilizadas ao longo de um programa, os módulos fornecem um meio de coletar conjuntos de funções (e como veremos nesta aula, tipos de dados personalizados) juntos para que possam ser usado por qualquer número de programas. ## Pacotes e MódulosUm módulo Python, de maneira sucinta, é um arquivo .py. Um módulo pode conter qualquer código Python que nós gostamos. Todos os programas que escrevemos até agora foram contidos em um único arquivo .py e, portanto, são módulos e programas. A principal diferença é que os programas são projetados para serem executados, enquanto os módulos são projetados para serem importados e usados por programas. Um pacote é simplesmente um diretório que contém um conjunto de módulos que podem estar inter-relacionados ou não. Os pacotes funcionam como coleções para organizar módulos de forma hierárquica. Quando queremos usar uma função de um módulo ou pacote, precisamos importá-lo. Existem muitas formas diferentes de importar módulos e pacotes. A sintaxe mais básica é:<jupyter_code>import numpy<jupyter_output><empty_output><jupyter_text>após o qual qualquer função em `numpy` pode ser chamada como `numpy.function()`. Você pode mudar internamente o nome do pacote, por exemplo porque ele é longo ou porque este nome conflita com o nome de um objeto seu. No exemplo abaixo nos alteramos o nome de numpy para np, o que é bastante padrão na computação científica:<jupyter_code>import numpy as np<jupyter_output><empty_output><jupyter_text>Todas as funções em `numpy` podem ser chamadas internamente como `np.function()`. Os pacotes também podem ter subpacotes. Por exemplo, o pacote `numpy` tem um subpacote chamado `random`, que tem um conjunto de funções para lidar com variáveis aleatórias. Se o pacote numpy for importado com `import numpy as np`, as funções no sub-pacote `random` podem ser chamadas como `np.random.function()`.## Um pouco sobre pacotesPython é uma ótima linguagem de programação de propósito geral por conta própria, mas com a ajuda de algums pacotes populares (numpy, scipy, matplotlib) torna-se um ambiente poderoso para a computação científica. `Numpy`, que significa Numerical Python, é a principal biblioteca usada na computação científica em Python. Ela fornece um objeto de vetores multidimensionais (matrizes) de alto desempenho e ferramentas para trabalhar com esses vetores.Se você precisar apenas de uma função específica, não é necessário importar todo o pacote. Por exemplo, se você quiser usar apenas a função coseno do pacote `numpy`, você pode importá-lo a partir de `numpy` assim:<jupyter_code>from numpy import cos<jupyter_output><empty_output><jupyter_text>após o qual você pode simplesmente chamar a função coseno como `cos()`. Você pode mesmo renomear as funções quando as importa. Por exemplo:<jupyter_code>from numpy import cos as coseno<jupyter_output><empty_output><jupyter_text>após o qual você pode chamar a função `coseno()` para calcular o coseno (eu sei, muito bobo, mas isso pode se tornar útil).Onde devemos inserir as declarações de importação? É prática comum colocar todas as instruções de importação no início dos arquivos .py, após a documentação do módulo. Além disso, recomendamos importar os módulos de biblioteca padrão do Python primeiro, depois módulos de biblioteca de terceiros e, finalmente, nossos próprios módulos.## Localizando MódulosQuando importamos pela primeira vez um módulo, se ele não for uma \textit{builtin}, o Python procura o módulo em cada diretório listado na variável de ambiente PYTHONPATH (`sys.path`) que normalmente inclui a pasta corrente em primeiro lugar. Uma conseqüência disso é que se criarmos um módulo ou programa com o mesmo nome de um dos módulos de biblioteca do Python, o nosso será encontrado primeiro, causando inevitavelmente problemas. Para evitar isso, nunca crie um programa ou módulo com o mesmo nome de um dos diretórios (pacotes) ou módulos de nível superior da biblioteca Python (i.e., não inclui os sub-módulos), a menos que você esteja fornecendo sua própria implementação desse módulo e que o substitua deliberadamente.Nós podemos incluir o diretório ou diretórios de nossos módulos na variável de ambiente através da biblioteca `sys.path` assim: <jupyter_code>import sys sys.path.append('my_path') import my_lib<jupyter_output><empty_output><jupyter_text># Módulos personalizadosNós abordamos até agora algumas ferramentas de software na resolução de problemas computacionais. A mais importante dessas ferramentas são os mecanismos de abstração para simplificar os projetos e controlar a complexidade das soluções. Os mecanismos de abstração incluem funções, módulos, objetos e classes. Em vários momentos começamos com uma visão externa de um recurso, mostrando o que ele faz e como ele pode ser usado. Por exemplo, para usar a função `sqrt` da \textit{builtin} `math`, você o importa: ``` from math import sqrt```, executa: ``` help(sqrt) ``` para aprender a usar a função corretamente e então inclua-o apropriadamente em seu código. Os mesmos procedimentos são seguidos para estruturas de dados \textit{builtin}, como strings e listas. Do ponto de vista de um usuário, você não teve que se preocupar com a forma como um recurso é executado. A beleza e a utilidade de uma abstração é que ela liberta você da necessidade de se preocupar com tais detalhes.Infelizmente, nem todas as abstrações são encontradas em buitins. Às vezes, você precisa personalizar uma abstração para atender às necessidades de um aplicativo especializado ou conjunto de aplicativos que você está desenvolvendo. Ao projetar sua própria abstração, você deve ter uma visão diferente da dos usuários e se preocupar com o funcionamento interno de um recurso. O programador que define uma nova função ou constrói um novo módulo de recursos está usando os recursos fornecidos por outros para criar novos componentes de software. Nós vamos mostrar nesta aula como projetar, implementar e testar outro mecanismo de abstração - uma classe. As linguagens de programação que permitem ao programador definir novas classes de objetos são chamados de linguagem orientada a objetos. Essas linguagens também suportam um estilo de programação chamado programação orientada a objetos. Ao contrário da programação baseada em objetos, que simplesmente usa objetos e classes prontas dentro uma estrutura de funções e código algorítmico, a programação orientada a objetos oferece mecanismos para conceber e construir sistemas de software inteiros a partir de classes cooperantes.Como funções, os objetos são abstrações. Uma função agrupa um algoritmo em uma única operação que pode ser chamada pelo nome. Um objeto envolve um conjunto de valores de dados -- seu estado, e um conjunto de operações -- seus métodos, em uma única entidade que pode ser referenciada com um nome. Isso torna um objeto uma abstração mais complexa do que uma função. Uma definição de classe é como um modelo para cada um dos objetos dessa classe. Este modelo contém: - Definições de todos os métodos que seus objetos reconhecem; - Descrições das estruturas de dados usadas para manter o estado de um objeto, ou seus atributos. Para ilustrar estas idéias, analisaremos a construção de um tipo abstrato de dados personalizado.## O módulo vector.pyO que queremos construir é um módulo que representa um tipo abstrato de dados chamado de vetor euclidiano (também conhecido como vetor geométrico) e operações sobre este. Um vetor é definido como uma entidade que possui tanto a magnitude quanto a direção. O vetor é tipicamente desenhado como uma seta; a direção é indicada por onde a flecha está apontando, e a magnitude pelo comprimento da própria seta. Por questão de simplicidade, vamos trabalhar com vetores de duas dimensões (representados num plano) embora os exemplos podem ser facilmente extendidos para três dimensões.Uma outra maneira de pensar em um vetor é a diferença entre dois pontos. Considere como você pode fazer instruções para andar de um ponto para outro. Aqui estão alguns vetores e possíveis traduções:![vector2](img/vector2.png)Portanto, podemos representar os vetores assim: <jupyter_code>vector(3, 4) vector(2, -1) vector(-15, 3)<jupyter_output><empty_output><jupyter_text>Antes de continuar a olhar para a tipo `vector` e seus métodos de soma, subtração etc., vamos recordar as principais operações sobre o tipo `vector` usando a notação encontrada em livros de matemática e física.### Vector AddDigamos que eu tenho os seguintes dois vetores:![vector3](img/vector3.png)Cada vetor tem dois componentes, $x$ e $y$. Para adicionar dois vetores juntos, simplesmente adicionamos ambos $x$ e os dois $y$:![vector4](img/vector4.png)### Vector Subtraction![vector5](img/vector5.png)### Vector Multiplication### Vector DivisionA divisão funciona como a multiplicação - simplesmente substituímos o sinal de multiplicação ($*$) pelo sinal de divisão ($/$).![vector7](img/vector7.png)### Vector MagnitudeA multiplicação e a divisão, como acabamos de ver, são meios pelos quais o comprimento do vetor pode ser alterado sem afetar a direção. Talvez você esteja se perguntando: "OK, então, como eu sei qual é o comprimento de um vetor? Conheço os componentes (x e y), mas qual o comprimento (em pixels) é a seta real? "Compreender como calcular o comprimento (também conhecido como magnitude) de um vetor é incrivelmente útil e importante.Observe nos diagramas anteriores como o vetor, desenhado como uma seta e dois componentes (x e y), cria um triângulo reto. Os lados são os componentes e a hipotenusa é a própria flecha. Armado com o theorema de Pitágoras podemos calcular a magnitude de v da seguinte maneira:$|v| = \sqrt{x^2 + y^2}$### Vector NormalizationPara normalizar um vetor, simplesmente dividimos cada componente por sua magnitude. Isso é bastante intuitivo. Digamos que um vetor tem comprimento 5. Bem, 5 dividido por 5 é 1. Então, olhando para o nosso triângulo retângulo, então precisamos escalar a hipotenusa para baixo dividindo-se por 5. Nesse processo os lados diminuirão, divididos por 5 também.![vector8](img/vector8.png)Existem muitas operações matemáticas que são comumente usadas com vetores, mas para nosso exemplo, vamos parar por aqui. Eis nosso código:<jupyter_code>#!/usr/bin/env python3 """vector.py: A simple little Vector class. Enabling basic vector math. """ __author__ = "Marcio Pereira" __license__ = "GPL" __version__ = "1.0" class Vector: """Represents a Vector.""" def __init__(self, x, y): """Constructor creates a Vector with x-axes and y-axes values.""" self.x = x self.y = y def __str__(self): """ Return a string representation of self.""" return "Vector(" + str(self.x) + ", " + str(self.y) + ")" # for testing: create and use some Vector objects if __name__ == '__main__': v1 = Vector(3, 4) v2 = Vector(2, -1) v3 = Vector(-15, 3) print(v1) print(v2.x) print(v3.y)<jupyter_output><empty_output><jupyter_text>A estrutura deste módulo (e a maioria dos outros) difere pouco do de um programa. A primeira linha é a linha `shebang` e, em seguida, é comum ter uma seqüência de citações entre aspas triplas que fornece uma visão geral do conteúdo do módulo, muitas vezes incluindo alguns exemplos de uso - esta é a `docstring` do módulo. Depois, temos alguns comentários (normalmente, os direitos autorais e as informações da licença), seguido do corpo do módulo e ao final, dentro de uma sentença `if`, uma parte reservada aos testes de suas funcionalidades.Sempre que um módulo é importado, o Python cria uma variável para o módulo chamado `__name__` e armazena o nome do módulo nesta variável. O nome de um módulo é simplesmente o nome do seu arquivo .py, mas sem a extensão. Então, neste exemplo, quando o módulo for importado `__name__` terá o valor `"vector"`, e a condição `if` não será atendida, então as últimas linhas de teste do módulo não serão executadas. Isto significa que estas últimas linhas têm praticamente nenhum custo quando o módulo é importado. Sempre que um arquivo .py é executado, o Python cria uma variável para o programa chamado `__name__` e define-o para a string `"__main__"`. Então, se fosse executar o `vector.py` como se fosse um programa, o Python configurará `__name__` para `"__main__"` e a condição `if` irá avaliar para `True` e as últimas linhas de teste do módulo serão executadas.## A Classe VectorA sintaxe de definição de classe tem duas partes: um cabeçalho de classe e um conjunto de definições de método que seguem o cabeçalho da classe. O cabeçalho da classe consiste no nome da classe e, no nome da classe "pai", onde este objeto "herda" suas caracteristicas. Quando não fornecido, esta classe é a classe `Object` de Python. Este mecanismo é conhecido como mecanismo de herança de classes, porém está fora do escopo deste curso. O nome da classe é um identificador em Python. Embora os nomes de tipos \textit{builtins} não estejam capitalizados, os programadores Python tipicamente usam seus nomes de classes capitalizados para distingui-los de nomes de variáveis.## Definições de MétodosTodas as definições do método são recuadas abaixo do cabeçalho da classe. Como os métodos são parecidos com funções, a sintaxe de suas definições é semelhante. Note, no entanto, que cada definição de método deve incluir um primeiro parâmetro chamado `self`, mesmo que esse método não receba nenhum argumento quando chamado. Quando um método é chamado com um objeto, o interpretador liga o parâmetro `self` a esse objeto para que o código do método possa se referir ao objeto pelo nome.### O Método `__init__`:A maioria das classes inclui um método especial chamado `__init__`. Observe que `__init__` deve começar e terminar com dois sublinhados consecutivos. Esse método também é chamado de construtor da classe, porque é executado automaticamente quando um usuário instancia a classe. Assim, quando o segmento de código: ``` v1 = Vector(3, 4) ``` é executado, o Python executa automaticamente o construtor ou o método `__init__` da classe Vector. O objetivo do construtor é inicializar os atributos de um objeto individual. Além de si próprio, o construtor Vector espera dois argumentos que fornecem os valores iniciais para esses atributos. A partir deste momento, quando nos referimos ao construtor de uma classe, queremos dizer seu método `__init__`.Os atributos de um objeto são representados como variáveis "de instância". Cada objeto individual possui seu próprio conjunto de variáveis "de instância". Essas variáveis servem de armazenamento para seu estado. O escopo de uma variável de instância (incluindo `self`) é a definição de classe inteira. Assim, todos os métodos da classe estão em posição de fazer referência às variáveis "da instância". A vida útil de uma variável de instância é a vida útil do objeto em si. Dentro da definição da classe, os nomes das variáveis "de instância" devem começar com `self`. Variáveis declaradas como da classe (i.e., sem `self`) são compartilhadas por todos os objetos daquele tipo.### O Método `__str__`:As classes geralmente incluem um método `__str__`. Esse método cria e retorna uma representação em string do estado de um objeto. Quando a função `str` é chamada com um objeto, o método `__str__` desse objeto é automaticamente invocado para obter a seqüência de caracteres que descreve o objeto. Por exemplo, a chamada da função `str(v1)` é equivalente à chamada do método `v1.__str__()`, e é mais simples de escrever. Note também que a função `print`do Python chama o método `__str__` da classe do objeto a ser impresso.<jupyter_code>print(v2)<jupyter_output>Vector(2, -1) <jupyter_text>### Outros métodos especiais:Podemos também definir as operações básicas de soma, subtração, multiplicação e divisão,usando métodos especiais (\textit{dunder methods}) do Python: <jupyter_code>class Vector: """Represents a Vector.""" def __init__(self, x, y): """Constructor creates a Vector with x-axes and y-axes values.""" self.x = x self.y = y def __str__(self): """ Return a string representation of self.""" return "Vector(" + str(self.x) + ", " + str(self.y) + ")" def __add__(self, other): """ Return the sum of self and Vector object other.""" return Vector(self.x + other.x, self.y + other.y) def __sub__(self, other): """ Return the difference of self and Vector object other.""" return Vector(self.x - other.x, self.y - other.y) def __mul__(self, alpha): """ Return the product of self and numeric object alpha.""" return Vector(self.x * alpha, self.y * alpha) def __rmul__(self, alpha): """ Return the reverse product of self and numeric object alpha.""" return Vector(self.x * alpha, self.y * alpha) def __truediv__(self, alpha): """ Return the division of self and numeric object alpha.""" return Vector(self.x / alpha, self.y / alpha) # for testing: create and use some Vector objects if __name__ == '__main__': v1 = Vector(3, 4) v2 = Vector(2, -1) v3 = Vector(-15, 3) print(v1) print(v2 + v1) print(v1 - v3) print(v3 * 2) print(4 * v1) print(v2 / 2)<jupyter_output><empty_output><jupyter_text>### Métodos normais de uma classeOs métodos "normais" de uma classe são chamados pelos nomes. Vamos definir as operações de Normalização e Magnitude de vetores como métodos comuns, isto é, sem recorrer aos métodos especiais:<jupyter_code>from math import sqrt class Vector: def __init__(self, x, y): """Constructor creates a Vector with x-axes and y-axes values.""" self.x = x self.y = y def __str__(self): """ Return a string representation of self.""" return "Vector(" + str(self.x) + ", " + str(self.y) + ")" def __add__(self, other): """ Return the sum of self and Vector object other.""" return Vector(self.x + other.x, self.y + other.y) def __sub__(self, other): """ Return the difference of self and Vector object other.""" return Vector(self.x - other.x, self.y - other.y) def __mul__(self, alpha): """ Return the product of self and numeric object alpha.""" return Vector(self.x * alpha, self.y * alpha) def __rmul__(self, alpha): """ Return the reverse product of self and numeric object alpha.""" return Vector(self.x * alpha, self.y * alpha) def __truediv__(self, alpha): """ Return the division of self and numeric object alpha.""" return Vector(self.x / alpha, self.y / alpha) def magnitude(self): """ Return the magnitude of self object.""" return sqrt(self.x ** 2 + self.y ** 2) def norm(self): """ Return the self object normalized.""" return self / self.magnitude() # for testing: create and use some Vector objects if __name__ == '__main__': v1 = Vector(3, 4) v2 = Vector(2, -1) v3 = Vector(-15, 3) print(v1.magnitude()) print(v1.norm())<jupyter_output><empty_output>
non_permissive
/17.Pacotes e Módulos.ipynb
zuj2019/mc102
10
<jupyter_start><jupyter_text># create new fields adm_df["year_term"] = ( adm_df["ACADEMIC_YEAR"] + "." + adm_df["ACADEMIC_TERM"].str.title() ) # week_number = ( # lambda r: (r["create_date"].date().isocalendar()[1]) # if (r["create_date"].date() >= date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1)) # else (date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1).isocalendar()[1]) # ) # adm_df["Week_Number"] = adm_df.apply(week_number, axis=1) print(adm_df.shape) adm_df.head(3)<jupyter_code># create new fields adm_df["year_term"] = ( adm_df["ACADEMIC_YEAR"] + "." + adm_df["ACADEMIC_TERM"].str.title() ) week_number = ( lambda r: (r["create_date"].date().isocalendar()[1]) if (r["create_date"].date() >= date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1)) else (date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1).isocalendar()[1]) ) adm_df["Week_Number"] = adm_df.apply(week_number, axis=1) print(adm_df.shape) adm_df.head(3) # convert ACADEMIC_YEAR to numeric keep numeric-valued records adm_df["ACADEMIC_YEAR"] = pd.to_numeric( adm_df["ACADEMIC_YEAR"], errors="coerce", downcast="integer" ) adm_df = adm_df.loc[adm_df["ACADEMIC_YEAR"].notnull()] print(adm_df.shape) adm_df.head(3) adm_df.loc[(adm_df['PEOPLE_CODE_ID']=='P000024505') & (adm_df['year_term']=='2012.Fall')]<jupyter_output><empty_output><jupyter_text>adm_week_number = ( lambda r: ( int( (pd.to_datetime(r["create_date"]) - pd.to_datetime(date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1))) / np.timedelta64(1,'W') ) ) if ( pd.to_datetime(r["create_date"]) >= pd.to_datetime(date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1)) ) else 0 ) adm_df["Admissions_Week"] = adm_df.apply(adm_week_number, axis=1) print(adm_df.shape) adm_df.head(3)<jupyter_code>adm_week_number = ( lambda r: ( r["Week_Number"] - (date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1).isocalendar()[1]) + 1 ) if ( r["Week_Number"] >= (date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1).isocalendar()[1]) ) else ( 53 + r["Week_Number"] - (date((int(r["ACADEMIC_YEAR"]) - 1), 9, 1).isocalendar()[1]) ) ) adm_df["Admissions_Week"] = adm_df.apply(adm_week_number, axis=1) print(adm_df.shape) adm_df.head(3) adm_df.loc[(adm_df['PEOPLE_CODE_ID']=='P000024505') & (adm_df['year_term']=='2012.Fall')] adm_keep_values = [ "300", "ACC", "ACXL", "CANC", "DEF", "DEFR", "DENY", "DPAC", "TRDP", "TRPD", "TRNS", "WAIT", "500", "PEND", "COMP", ] adm_keep_cols = ["PEOPLE_CODE_ID", "year_term", "Admissions_Week", "field_value"] adm_df = adm_df.loc[(adm_df["field_value"].isin(adm_keep_values)), adm_keep_cols] # admissions status table admission_status = { "300": "Applied", "ACC": "Accepted", "ACXL": "Canceled", "CANC": "Canceled", "DEF": "Canceled", "DEFR": "Canceled", "DENY": "Canceled", "DPAC": "Deposited", "TRDP": "Deposited", "TRPD": "Deposited", "TRNS": "Accepted", "WAIT": "Accepted", "500": "Deposited", "PEND": "Applied", "COMP": "Applied", } adm_stat = pd.DataFrame( list(admission_status.items()), columns=["field_value", "admission_status"] ) adm_df1 = ( pd.merge(adm_df, adm_stat, on=["field_value"], how="left") .drop(["field_value"], axis=1) .drop_duplicates( ["PEOPLE_CODE_ID", "year_term", "Admissions_Week", "admission_status"] ) ) print(adm_df1.shape) adm_df1 = adm_df1.sort_values( ["year_term", "PEOPLE_CODE_ID", "admission_status", "Admissions_Week"] ).drop_duplicates(["year_term", "PEOPLE_CODE_ID", "admission_status"], keep="first") print(adm_df1.shape) adm_df1.loc[(adm_df1['PEOPLE_CODE_ID']=='P000024505')] e = adm_df1.pivot_table( index=["year_term", "PEOPLE_CODE_ID"], columns=["admission_status"], values=["Admissions_Week"], ) e = e.fillna(np.int(54)) print(e.shape) e.head() e.info() e.loc[('2012.Fall', 'P000024505')] # function returns status for week def f_status(field, data_frame, n): f_week = ( lambda df: 1 if ( (df[("Admissions_Week", field)] <= n) & (df[("Admissions_Week", "Canceled")] > n) ) else 0 ) return data_frame.apply(f_week, axis=1) # function returns DataFrame of 53 week status values def fill_weeks(field, data_frame): weeks = range(0, 54) r = pd.DataFrame( np.zeros((data_frame.shape[0], 54)), index=data_frame.index, columns=[f"{w:02d}" for w in weeks], ) for w in weeks: f = f"{w:02d}" r.loc[:, f] = f_status(field, data_frame, w) r.loc[:, "stage"] = field r = r.reset_index().set_index(["year_term", "stage", "PEOPLE_CODE_ID"]) return r stage_list = ["Applied", "Accepted", "Deposited"] w = pd.DataFrame() for stg in stage_list: w = pd.concat([w, fill_weeks(stg, e)]) print(w.shape) w.head() w.info() w.loc[('2018.Fall', 'Accepted',)] # add CURRICULUM field sql_str = ( "SELECT PEOPLE_CODE_ID, ACADEMIC_YEAR, ACADEMIC_TERM, " + "ACADEMIC_SESSION, CURRICULUM, PRIMARY_FLAG " + "FROM ACADEMIC WHERE " + "PRIMARY_FLAG = 'Y' AND " + f"ACADEMIC_YEAR >= '{begin_year}' " ) curriculum_df = pd.read_sql_query(sql_str, connection) curriculum_df["year_term"] = ( curriculum_df["ACADEMIC_YEAR"] + "." + curriculum_df["ACADEMIC_TERM"].str.title() ) curriculum_df = curriculum_df.rename(columns={"CURRICULUM": "curriculum"}) curr_flds = ["PEOPLE_CODE_ID", "year_term", "curriculum"] curriculum_df = curriculum_df[curr_flds] curriculum_df = curriculum_df.drop_duplicates(curr_flds) y = pd.merge( w.reset_index(), curriculum_df, on=["year_term", "PEOPLE_CODE_ID"], how="left" ) print(y.shape) y.head() y.to_hdf(data_store, key="weekly", mode="w", data_columns=True, complevel=0) <jupyter_output>(45015, 58)
non_permissive
/funnel/admissions.py_Test_20201217.ipynb
JeffWalton-PSC/Admissions
2
<jupyter_start><jupyter_text>### A Fredkin gate is be obtained with very good fidelity, $\simeq 99.999\%$, starting with all interactions on, and only $\sigma_z$ operators as self-interactions.<jupyter_code>net = QubitNetwork( num_qubits=4, # interactions=('all', ['xy', 'xx', 'yy', 'zz']), interactions='all', self_interactions=('all', ['z']), system_qubits=[0, 1, 2] # J=new_Jvalues ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=10, target_gate=qutip.fredkin(), training_dataset_size=100, test_dataset_size=1000, decay_rate=.01, saveafter_file='nets/fredkin_best.pickle' )<jupyter_output>Generating training data... Building the model... Let's roll! <jupyter_text>## Failing to implement 3-qubit QFT with 1 ancilla<jupyter_code>net = QubitNetwork( num_qubits=4, interactions='all', system_qubits=[0, 1, 2] ) sgd_optimization( net=net, learning_rate=1, n_epochs=400, batch_size=2, target_gate=qutip.qip.algorithms.qft.qft(3), training_dataset_size=100, test_dataset_size=1000, decay_rate=.01, saveafter_file='nets/qft.pickle' )<jupyter_output>Generating training data... Building the model... Let's roll! <jupyter_text>## Success in implementing Hadamard gate with 3 qubits + 1 ancilla<jupyter_code>net = QubitNetwork( num_qubits=4, interactions='all', system_qubits=3 ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=qutip.hadamard_transform(3), training_dataset_size=100, test_dataset_size=1000, decay_rate=.01, saveafter_file='nets/hadamard.pickle' ) Jvalues = net.J.get_value() vanishing_indices = np.where(np.abs(Jvalues) < 1e-4)[0] [net.J_index_to_interaction(v) for v in vanishing_indices]<jupyter_output><empty_output><jupyter_text>### Unitaries that are tensor product of 1-qubit gates are reachable (of course)<jupyter_code>rand_U = qutip.tensor([qutip.rand_unitary_haar(2) for _ in range(3)]) net = QubitNetwork( num_qubits=4, interactions='all', system_qubits=3 ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=rand_U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 # saveafter_file='nets/hadamard.pickle' ) p00 = qutip.ket2dm(qutip.basis(2, 0)) p11 = qutip.ket2dm(qutip.basis(2, 1)) u1 = qutip.rand_unitary_haar(4) u2 = qutip.rand_unitary_haar(4) u1.dims = u2.dims = [[2, 2]] * 2 U = qutip.tensor(p00, u1) + qutip.tensor(p11, u2) U net = QubitNetwork( num_qubits=4, interactions='all', system_qubits=3 ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01, saveafter_file='nets/controlled_stuff.pickle' ) U = qutip.tensor(qutip.sigmax(), qutip.sigmax(), qutip.qeye(2)) net = QubitNetwork( num_qubits=3, system_qubits=3, interactions=('all', ['xx', 'zz', 'z']), ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) gate = qutip.Qobj(net.get_current_gate()) v = net.J.get_value() display(v) [net.J_index_to_interaction(i) for i in np.where(np.abs(v) > 1)[0]] U = qutip.tensor(qutip.sigmax(), qutip.sigmax()) net = QubitNetwork( num_qubits=2, system_qubits=2, interactions=('all', ['xx', 'yy', 'zz', 'z']), ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) net.J.get_value() p0 = qutip.Qobj([[1, 0], [0, 0]]) p1 = qutip.Qobj([[0, 0], [0, 1]]) i2 = qutip.qeye(2) u2 = qutip.rand_unitary_haar(2) u2p = qutip.rand_unitary_haar(2) U = (qutip.tensor(p0, p0, i2) + qutip.tensor(p0, p1, i2) + qutip.tensor(p1, p0, i2) + qutip.tensor(p1, p1, u2)) net = QubitNetwork( num_qubits=4, system_qubits=3, interactions='all', ancillae_state=qutip.basis(2, 0) ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) p0 = qutip.Qobj([[1, 0], [0, 0]]) p1 = qutip.Qobj([[0, 0], [0, 1]]) i2 = qutip.qeye(2) u2 = qutip.rand_unitary_haar(2) u2p = qutip.rand_unitary_haar(2) U = (qutip.tensor(p0, p0, u2p) + qutip.tensor(p0, p1, i2) + qutip.tensor(p1, p0, i2) + qutip.tensor(p1, p1, u2)) net = QubitNetwork( num_qubits=4, system_qubits=3, interactions='all', ancillae_state=qutip.basis(2, 0) ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) p0 = qutip.Qobj([[1, 0], [0, 0]]) p1 = qutip.Qobj([[0, 0], [0, 1]]) i2 = qutip.qeye(2) u2 = qutip.rand_unitary_haar(2) u2p = qutip.rand_unitary_haar(2) u2pp = qutip.rand_unitary_haar(2) u2ppp = qutip.rand_unitary_haar(2) U = (qutip.tensor(p0, p0, u2p) + qutip.tensor(p0, p1, u2pp) + qutip.tensor(p1, p0, u2ppp) + qutip.tensor(p1, p1, u2)) net = QubitNetwork( num_qubits=4, system_qubits=3, interactions='all', ancillae_state=qutip.basis(2, 0) ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) p0 = qutip.Qobj([[1, 0], [0, 0]]) p1 = qutip.Qobj([[0, 0], [0, 1]]) i2 = qutip.qeye(2) u2 = qutip.rand_unitary_haar(2) u2p = qutip.rand_unitary_haar(2) u2pp = qutip.rand_unitary_haar(2) u2ppp = qutip.rand_unitary_haar(2) U = (qutip.tensor(p0, p0, u2p) + qutip.tensor(p0, p1, i2) + qutip.tensor(p1, p0, i2) + qutip.tensor(p1, p1, u2)) net = QubitNetwork( num_qubits=4, system_qubits=3, interactions='all', ancillae_state=qutip.basis(2, 0) ) sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=2, target_gate=U, training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) np.savetxt('matrices/CCrandomrandom_4q1a_all_all.txt', net.get_current_gate(), delimiter=',') interactions = [((0, 1), 'zz'), ((0, 2), 'zz'), ((1, 2), 'zz'), ((1, 3), 'zz'), ((2, 3), 'zz'), ((0, 3), 'xx'), (0, 'z'), (1, 'z'), (2, 'z'), (3, 'z'), (0, 'x'), (1, 'x'), (2, 'x'), (3, 'x'), (0, 'y'), (1, 'y'), (2, 'y'), (3, 'y')] eta = 0.8182 xi = 0.0587 initial_ancilla = qutip.Qobj([[np.cos(eta)], [np.sin(eta)]]) net = QubitNetwork( num_qubits=4, system_qubits=3, interactions=interactions, ancillae_state=initial_ancilla ) net = sgd_optimization( net=net, learning_rate=1, n_epochs=1000, batch_size=5, target_gate=qutip.toffoli(), training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) interactions = [((0, 1), 'zz'), ((0, 2), 'zz'), ((1, 2), 'zz'), ((1, 3), 'zz'), ((2, 3), 'zz'), ((0, 3), 'xx'), (0, 'z'), (1, 'z'), (2, 'z'), (3, 'z'), (0, 'x'), (1, 'x'), (2, 'x'), (3, 'x')] eta = 0.8182 xi = 0.0587 initial_ancilla = qutip.Qobj([[np.cos(eta)], [np.sin(eta)]]) net1 = QubitNetwork( num_qubits=4, system_qubits=3, interactions=interactions, ancillae_state=initial_ancilla ) net1 = sgd_optimization( net=net1, learning_rate=1, n_epochs=1000, batch_size=5, target_gate=qutip.toffoli(), training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) interactions = [((0, 1), 'zz'), ((0, 2), 'zz'), ((1, 2), 'zz'), ((1, 3), 'zz'), ((2, 3), 'zz'), ((0, 3), 'xx'), (0, 'z'), (1, 'z'), (2, 'z'), (3, 'z'), (0, 'x'), (1, 'x'), (2, 'x'), (3, 'x')] eta = 0.8182 xi = 0.0587 initial_ancilla = qutip.Qobj([[np.cos(eta)], [np.sin(eta)]]) net2 = QubitNetwork( num_qubits=4, system_qubits=3, interactions=interactions, ancillae_state=initial_ancilla ) net2 = sgd_optimization( net=net2, learning_rate=1, n_epochs=1000, batch_size=5, target_gate=qutip.toffoli(), training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) interactions = [((0, 1), 'zz'), ((0, 2), 'zz'), ((1, 2), 'zz'), ((1, 3), 'zz'), ((2, 3), 'zz'), ((0, 3), 'xx'), (0, 'z'), (1, 'z'), (2, 'z'), (3, 'z'), (0, 'x'), (1, 'x'), (2, 'x'), (3, 'x')] eta = 0.8182 xi = 0.0587 initial_ancilla = qutip.Qobj([[np.cos(eta)], [np.sin(eta)]]) net3 = QubitNetwork( num_qubits=4, system_qubits=3, interactions=interactions, ancillae_state=initial_ancilla ) net3 = sgd_optimization( net=net3, learning_rate=1, n_epochs=1000, batch_size=5, target_gate=qutip.toffoli(), training_dataset_size=100, test_dataset_size=1000, decay_rate=.01 ) import scipy.sparse import scipy.linalg as la import sympy import theano import theano.tensor.slinalg import theano.tensor as T sympy.init_printing() n = 4 A = np.zeros(shape=(n, n), dtype=np.float) A += scipy.sparse.rand(4, 4, density=.8) A = np.asarray(A) # numerical differentiation x = 2. h = 0.001 numerical_diff = (la.expm((x + h) * A) - la.expm(x * A)) / h # automatic differentiation with theano x = T.dscalar('x') expA = T.slinalg.expm(x * A) expA_flat = T.flatten(expA) def compute_element_grad(idx, flattened_matrix): return T.grad(flattened_matrix[idx], wrt=x) g_x_flat, _ = theano.scan( fn=compute_element_grad, sequences=T.arange(expA_flat.shape[0]), non_sequences=[expA_flat] ) # deflatten result g_x = T.reshape(g_x_flat, newshape=expA.shape) # here is where the computational graph is actually compiled gradient = theano.function(inputs=[x], outputs=g_x) # compute and display gradient computed with AD display(gradient(2.)) # display gradient computed with numerical differentiation display(numerical_diff)<jupyter_output><empty_output>
no_license
/notebooks/quantum_gate_learning.ipynb
theneva/quantum-gate-learning
4
<jupyter_start><jupyter_text># AssignmentQ1. Write the NumPy program to create an array of ones and an array of zeros? Expected OutputCreate an array of zeros Default type is float [[ 0. 0.]] Type changes to int [[0 0]] Create an array of ones Default type is float [[ 1. 1.]] Type changes to int [[1 1]]<jupyter_code>import numpy as np print("Create an array of zeros") zeros_array = np.zeros((1,2)) print("Default type is float") print(zeros_array) print("Type changes to int") zeros_array = np.zeros((1,2), dtype = np.int) print(zeros_array) print("\nCreate an array of ones") ones_array= np.ones((1,2)) print("Default type is float") print(ones_array) print("Type changes to int") ones_array = np.ones((1,2), dtype = np.int) print(ones_array)<jupyter_output>Create an array of zeros Default type is float [[0. 0.]] Type changes to int [[0 0]] Create an array of ones Default type is float [[1. 1.]] Type changes to int [[1 1]] <jupyter_text>Q2. Write the NumPy program to change the dimension of an array? Expected Output6 rows and 0 columns (6,) (3, 3) -> 3 rows and 3 columns [[1 2 3] [4 5 6] [7 8 9]] Change array shape to (3, 3) -> 3 rows and 3 columns [[1 2 3] [4 5 6] [7 8 9]]<jupyter_code>import numpy as np array = np.array([1, 2, 3, 4, 5, 6]) print("6 rows and 0 columns") print(array.shape) ndarray = np.array([[1, 2, 3],[4, 5, 6],[7,8,9]]) print("\n(3, 3) -> 3 rows and 3 columns ") print(ndarray) array = np.array([1,2,3,4,5,6,7,8,9]) print("\nChange array shape from (6,) to (3, 3) -> 3 rows and 3 columns ") array.shape = (3, 3) print(array)<jupyter_output>6 rows and 0 columns (6,) (3, 3) -> 3 rows and 3 columns [[1 2 3] [4 5 6] [7 8 9]] Change array shape from (6,) to (3, 3) -> 3 rows and 3 columns [[1 2 3] [4 5 6] [7 8 9]] <jupyter_text>Q3. Write the NumPy program to create a new shape to an array without changing its data ? Reshape 3x2- [[1 2] [3 4] [5 6]] Reshape 2x3- [[1 2 3] [4 5 6]]<jupyter_code>import numpy as np x = np.array([1, 2, 3, 4, 5, 6]) y = np.reshape(x,(3,2)) print("Reshape 3x2 -") print(y) z = np.reshape(x,(2,3)) print("\nReshape 2x3 -") print(z)<jupyter_output>Reshape 3x2 - [[1 2] [3 4] [5 6]] Reshape 2x3 - [[1 2 3] [4 5 6]] <jupyter_text>Q4. Write the NumPy program to create a new array of 3*5, filled with 2? Expected Output- [[2 2 2 2 2] [2 2 2 2 2] [2 2 2 2 2]] [[2 2 2 2 2] [2 2 2 2 2] [2 2 2 2 2]]<jupyter_code>import numpy as np #Method-1 using no.full print("Method-1 using no.full") full_array = np.full((3, 5), 2, dtype=np.uint) print(full_array) #Method-2 using no.ones print("\nMethod-2 using no.ones") ones_array = np.ones([3, 5], dtype=np.uint) *2 print(ones_array)<jupyter_output>Method-1 using no.full [[2 2 2 2 2] [2 2 2 2 2] [2 2 2 2 2]] Method-2 using no.ones [[2 2 2 2 2] [2 2 2 2 2] [2 2 2 2 2]] <jupyter_text>Q5. Write the NumPy program to create a 3-D array with ones on a diagonal and zeros elsewhere? Expected Output- [[ 1. 0. 0.] [ 0. 1. 0.] [ 0. 0. 1.]]<jupyter_code>import numpy as np nd_array = np.eye(3) print(nd_array)<jupyter_output>[[1. 0. 0.] [0. 1. 0.] [0. 0. 1.]] <jupyter_text>Q6. Write the NumPy program to split an array of 14 elements into the 3 arrays and each of which has 2, 4, and 8 elements in original order? Expected OutputOriginal array- [ 1 2 3 4 5 6 7 8 9 10 11 12 13 14] After splitting- [array([1, 2]), array([3, 4, 5, 6]), array([ 7, 8, 9, 10, 11, 12, 13, 14])]<jupyter_code>import numpy as np array = np.arange(1, 15) print("Original array- ", array) print("\nAfter splitting- ") print(np.split(array, [2, 6]))<jupyter_output>Original array- [ 1 2 3 4 5 6 7 8 9 10 11 12 13 14] After splitting- [array([1, 2]), array([3, 4, 5, 6]), array([ 7, 8, 9, 10, 11, 12, 13, 14])] <jupyter_text>Q7. Write the NumPy program to split of an array of shape 4x4 it into two arrays along the second axis ? Sample array - [[ 0 1 2 3] [ 4 5 6 7] [ 8 9 10 11] [12 13 14 15]] Expected Output- [array([[ 0, 1], [ 4, 5], [ 8, 9], [12, 13]]), array([[ 2, 3], [ 6, 7], [10, 11], [14, 15]]), array([], shape=(4, 0), dtype=int64)]<jupyter_code>import numpy as np array = np.arange(16).reshape((4, 4)) print("Original array:\n", array) ''' Default dtype allotted is 32 bit ''' print("\nOutput after splitting horizontally:") print(np.hsplit(array, [2, 6]))<jupyter_output>Original array: [[ 0 1 2 3] [ 4 5 6 7] [ 8 9 10 11] [12 13 14 15]] Output after splitting horizontally: [array([[ 0, 1], [ 4, 5], [ 8, 9], [12, 13]]), array([[ 2, 3], [ 6, 7], [10, 11], [14, 15]]), array([], shape=(4, 0), dtype=int32)] <jupyter_text>Q8. Write the NumPy program to create a 5x5 matrix with row values ranging from 0 to 4? Original array- [[ 0. 0. 0. 0. 0.] [ 0. 0. 0. 0. 0.] [ 0. 0. 0. 0. 0.] [ 0. 0. 0. 0. 0.] [ 0. 0. 0. 0. 0.]] Row values ranging from 0 to 4. [[ 0. 1. 2. 3. 4.] [ 0. 1. 2. 3. 4.] [ 0. 1. 2. 3. 4.] [ 0. 1. 2. 3. 4.] [ 0. 1. 2. 3. 4.]]<jupyter_code>import numpy as np array = np.zeros((5,5)) print("Original array -") print(array) print("\nRow values ranging from 0 to 4.") array += np.arange(5) print(array)<jupyter_output>Original array - [[0. 0. 0. 0. 0.] [0. 0. 0. 0. 0.] [0. 0. 0. 0. 0.] [0. 0. 0. 0. 0.] [0. 0. 0. 0. 0.]] Row values ranging from 0 to 4. [[0. 1. 2. 3. 4.] [0. 1. 2. 3. 4.] [0. 1. 2. 3. 4.] [0. 1. 2. 3. 4.] [0. 1. 2. 3. 4.]] <jupyter_text>Q9. Write the NumPy program to create an array of zeros and three column types (integer, float, character)? Expected Output- [(1, 2., b'Albert Einstein') (2, 2., b'Edmond Halley') (3, 3., b'Gertrude B. Elion')]<jupyter_code>import numpy as np #data type a20 choosen as all the characters of names are less than 20 data = np.zeros((3,), dtype=('i4,f4,a20')) new_data = [(1, 2., "Albert Einstein"), (2, 2., "Edmond Halley"), (3, 3., "Gertrude B. Elion")] data[:] = new_data print("Output - \n", data)<jupyter_output>Output - [(1, 2., b'Albert Einstein') (2, 2., b'Edmond Halley') (3, 3., b'Gertrude B. Elion')] <jupyter_text>Q10. Write the NumPy program to remove the negative values in the numpy array with 0? Expected OutputOriginal array: [-1 -4 0 2 3 4 5 -6] Replace the negative values of the said array with 0- [0 0 0 2 3 4 5 0]<jupyter_code>import numpy as np array = np.array([-1, -4, 0, 2, 3, 4, 5, -6]) print("Original array:") print(array) print("\nReplace the negative values of the said array with 0 -") array[array < 0] = 0 print(array)<jupyter_output>Original array: [-1 -4 0 2 3 4 5 -6] Replace the negative values of the said array with 0 - [0 0 0 2 3 4 5 0] <jupyter_text>Q11. Write the NumPy program to compute the histogram of a set of data?<jupyter_code>import numpy as np import matplotlib.pyplot as plt plt.hist([0, 1, 1, 2, 2, 2, 3, 3, 4], bins= range(6)) plt.show()<jupyter_output><empty_output><jupyter_text>Q12. Write the NumPy program to compute the line graph of a set of data?<jupyter_code>import sys import numpy as np import matplotlib.pyplot as plt arr = np.random.randint(20, 80, 30) y, x = np.histogram(arr, bins=np.arange(101)) fig, ax = plt.subplots() ax.plot(x[:-1], y) fig.savefig('D:/Basha/New folder/ax.jpg')<jupyter_output><empty_output><jupyter_text>Q13. Write the NumPy program to extracts all the elements from second row from given (4x4) array? Sample OutputOriginal array- [[ 0 1 2 3] [ 4 5 6 7] [ 8 9 10 11] [12 13 14 15]] Extracted data- Second row [4 5 6 7]<jupyter_code>import numpy as np arra_data = np.arange(0,16).reshape((4, 4)) print("Original array -") print(arra_data) print("\nExtracted data - Second row") print(arra_data[1,:])<jupyter_output>Original array - [[ 0 1 2 3] [ 4 5 6 7] [ 8 9 10 11] [12 13 14 15]] Extracted data - Second row [4 5 6 7] <jupyter_text>Q14. Write the NumPy program to extract first element of the second row and fourth element of fourth row from a given (4x4) array? Sample OutputOriginal array- [[ 0 1 2 3] [ 4 5 6 7] [ 8 9 10 11] [12 13 14 15]] Extracted data- First element of the second row and fourth element of fourth row [ 4 15] <jupyter_code>import numpy as np arra_data = np.arange(0,16).reshape((4, 4)) print("Original array -") print(arra_data) print("\nExtracted data - First element of the second row and fourth element of fourth row ") print(arra_data[[1,3],[0,3]])<jupyter_output>Original array - [[ 0 1 2 3] [ 4 5 6 7] [ 8 9 10 11] [12 13 14 15]] Extracted data - First element of the second row and fourth element of fourth row [ 4 15] <jupyter_text>Q15. Write the NumPy program to add two arrays A and B of sizes (3,3) and (,3)? Sample OutputOriginal arrayArray-1 [[1 1 1] [1 1 1] [1 1 1]] Array-2 [0 1 2] A + B: [[1 2 3] [1 2 3] [1 2 3]]<jupyter_code>import numpy as np A = np.ones((3,3),dtype=np.int) B = np.arange(3) print("Original array:") print("\nArray-1") print(A) print("\nArray-2") print(B) print("\nAddition of two arrays A + B:") new_array = A + B print(new_array)<jupyter_output>Original array: Array-1 [[1 1 1] [1 1 1] [1 1 1]] Array-2 [0 1 2] Addition of two arrays A + B: [[1 2 3] [1 2 3] [1 2 3]] <jupyter_text>Q16. Write the NumPy program to copy data from a given array to another array? Sample OutputOriginal array- [24 27 30 29 18 14] Copy of the said array- [24 27 30 29 18 14]<jupyter_code>import numpy as np array1 = np.array([24, 27, 30, 29, 18, 14]) print("Original array:") print(array1) copy_array1 = np.empty_like (array1) copy_array1[:] = array1 print("\nCopy of the said array:") print(copy_array1)<jupyter_output>Original array: [24 27 30 29 18 14] Copy of the said array: [24 27 30 29 18 14] <jupyter_text>Q17. Write the NumPy program to calculate the sum of all columns of the 2D numpy array? Sample OutputOriginal array- [[ 0 1 2 3 4 5 6 7 8] [ 9 10 11 12 13 14 15 16 17] [18 19 20 21 22 23 24 25 26] [27 28 29 30 31 32 33 34 35]] Sum of all columns- [54 58 62 66 70 74 78 82 86]<jupyter_code>import numpy as np num = np.arange(36) array1 = np.reshape(num, [4, 9]) print("Original array -") print(array1) result = array1.sum(axis=0) print("\nSum of all columns -") print(result)<jupyter_output>Original array - [[ 0 1 2 3 4 5 6 7 8] [ 9 10 11 12 13 14 15 16 17] [18 19 20 21 22 23 24 25 26] [27 28 29 30 31 32 33 34 35]] Sum of all columns - [54 58 62 66 70 74 78 82 86] <jupyter_text>Q18. Write the NumPy program to calculate averages without NaNs along the given array? Sample OutputOriginal array- [[10. 20. 30.] [40. 50. nan] [nan 6. nan] [nan nan nan]] Averages without NaNs along the said array- [20. 45. 6. nan]<jupyter_code>import numpy as np array1 = np.array([[10, 20 ,30], [40, 50, np.nan], [np.nan, 6, np.nan], [np.nan, np.nan, np.nan]]) print("Original array -") print(array1) temp = np.ma.masked_array(array1,np.isnan(array1)) #masked nan values to compute an average of the remaining numbers result = np.mean(temp, axis=1) print("\nAverages without NaNs along the said array -") print(result.filled(np.nan))<jupyter_output>Original array - [[10. 20. 30.] [40. 50. nan] [nan 6. nan] [nan nan nan]] Averages without NaNs along the said array - [20. 45. 6. nan]
no_license
/DL - Subjective Assignment - 6 - Numpy 2.ipynb
lalubasha434/DL-Assignments
18
<jupyter_start><jupyter_text>### Crear base de datos con SQLite Primero vamos a crear una base de datos<jupyter_code>import sqlite3 conn = sqlite3.connect('todo.db') conn.execute("CREATE TABLE todo (id INTEGER PRIMARY KEY, task char(100) NOT NULL, status boll NOT NULL)") conn.execute("INSERT INTO todo (task, status)VALUES('Crear hola mundo en bottlepy', 1)") conn.execute("INSERT INTO todo (task, status)VALUES('Documentar su creacion en jupyter',1)") conn.execute("INSERT INTO todo (task, status)VALUES('Crar un proyecto nuevo de una lista ToDO',1)") conn.execute("INSERT INTO todo (task, status)VALUES('Documentar creacion de ToDo en jupyter',0)") conn.commit() import sqlite3 from bottle import Bottle, run app= Bottle() @app.route('/todo') @app.route('/') def todo_list(): connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("SELECT id, task FROM todo WHERE status LIKE '1'" ) result = cursor.fetchall() return str(result) run(app, host='0.0.0.0', port=5000, debug=True) import sqlite3 from bottle import Bottle, run, template app= Bottle() @app.route('/todo') @app.route('/') def todo_list(): connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("SELECT id, task FROM todo WHERE status LIKE '1'" ) result = cursor.fetchall() output = template('make_table', rows=result) return output run(app, host='0.0.0.0', port=5000, debug=True) import sqlite3 from bottle import Bottle, run, template, request app= Bottle() @app.route('/todo') @app.route('/') def todo_list(): connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("SELECT id, task FROM todo WHERE status LIKE '0'" ) result = cursor.fetchall() output = template('make_table', rows=result) return output @app.route('/new', method='GET') def new_item(): return template('new_task.tpl') @app.route('/new', method='POST') def new_item_do(): new_item = request.POST.task.strip() connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("INSERT INTO todo (task,status)VALUES (?,?)", (new_item, 0)) new_id = cursor.lastrowid connection.commit() cursor.close() return f""" <meta HTTP-EQUIV="REFRESH" content="5; url={request.urlparts[0]+'://'+request.urlparts[1]}/"> <p>The new task was insert into the database, the ID is {new_id}</p>""" run(app, host='0.0.0.0', port=5000, debug=True) import sqlite3 from bottle import Bottle, run, template, request app= Bottle() @app.route('/todo') @app.route('/') def todo_list(): connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("SELECT id, task FROM todo WHERE status LIKE '0'" ) result = cursor.fetchall() output = template('make_table', rows=result) return output @app.route('/new', method='GET') def new_item_do(): if request.GET.save: new_item = request.GET.task.strip() connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("INSERT INTO todo (task,status)VALUES (?,?)", (new_item, 0)) new_id = cursor.lastrowid connection.commit() cursor.close() return f""" <meta HTTP-EQUIV="REFRESH" content="5; url={request.urlparts[0]+'://'+request.urlparts[1]}/"> <p>The new task was insert into the database, the ID is {new_id}</p>""" else: return template('new_task.tpl') run(app, host='0.0.0.0', port=5000, debug=True)<jupyter_output><empty_output><jupyter_text>## Agregar una ruta para editar tareas<jupyter_code>import sqlite3 from bottle import Bottle, run, template, request app= Bottle() @app.route('/todo') @app.route('/') def todo_list(): connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("SELECT id, task FROM todo WHERE status LIKE '0'" ) result = cursor.fetchall() output = template('make_table', rows=result) return output @app.route('/new', method='GET') def new_item_do(): if request.GET.save: new_item = request.GET.task.strip() connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("INSERT INTO todo (task,status)VALUES (?,?)", (new_item, 0)) new_id = cursor.lastrowid connection.commit() cursor.close() return f""" <meta HTTP-EQUIV="REFRESH" content="5; url={request.urlparts[0]+'://'+request.urlparts[1]}/"> <p>The new task was insert into the database, the ID is {new_id}</p>""" else: return template('new_task.tpl') @app.route('/edit/<num:int>', method='GET') def edit_item(num): if request.GET.save: edit = request.GET.task.strip() status = request.GET.status.strip() if status == 'open': status = 0 else: status = 1 connection = sqlite3.connect('todo.db') cursor = connection.cursor() cursor.execute("UPDATE todo SET task = ?, status = ? WHERE id LIKE ?",(edit, status, num)) connection.commit() return f'<p>The item number {num} was successfully update</p>' else: connection=sqlite3.connect('todo.db') cursor=connection.cursor() cursor.execute("SELECT task FROM todo WHERE id LIKE ?",(str(num),)) cur_data=cursor.fetchone() return template('edit_task',old=cur_data,num=num) run(app, host='0.0.0.0', port=5000, debug=True)<jupyter_output><empty_output>
no_license
/TodoList-Bottle.ipynb
RamonAgramon/Jupyternotebooks
2
<jupyter_start><jupyter_text># Electronic structure through (quantum) annealing In this project we map the electronic structure Hamiltonian to an Ising Hamiltonian and find the ground state energy. Refer to the following references: [1] https://arxiv.org/abs/1706.00271 [2] https://arxiv.org/abs/1811.05256 [3] https://arxiv.org/abs/1208.5986 We use molecular Hydrogen $H_2$ as an example. Assuming the atomic nucleus does not move due to its larger mass, the Hamiltonian which governs the electronic state can be transformed to a qubit representation appropriate for simulation on a quantum computer [3]. See Ref. [2], Eq. (6) for the $n$ qubit Hamiltonian, which encodes the electronic structure problem. Following Ref. [1], we then encode this problem in a classical Ising model, appropriate for annealing. This requires $r$ ancillary bit for each $n$ qubit. The qubit Hamiltonian for moledular hydrogen $H_2$ is given by Eq. (37) in Ref. [1]. After the mapping described above, the problem eventually maps to the 2-local Ising-type Hamiltonian Eq. (41). This goal becomes the calculation of the ground state energy of this Hamiltonian. <jupyter_code>import numpy as np from scipy.special import logsumexp import pandas as pd def energy(spins, J, h): # J - 2D np.array # h - 1D np.array # spins - 1D np.array (entries +/- 1) interaction = 0.5 * np.einsum("...i,ij,...j->...", spins, J, spins) field = np.einsum("...i,i->...", spins, h) return interaction + field def energy_diff(i, spins, J, h): return -2 * np.einsum("j,...j->...", J[i, :], spins) * spins[..., i] - 2 * h[i] * spins[..., i] num_spins = 10 # random interaction+field ising model J = np.random.randn(num_spins, num_spins) J = (J + J.T) / 2 for i in range(J.shape[0]): J[i, i] = 0 h = np.random.randn(num_spins) spins = (2*np.random.randint(2, size=(num_spins,)) - 1) # standard classical ising with no field J = np.zeros((num_spins, num_spins)) for i in range(J.shape[0]): J[i, (i+1) % num_spins] = -1 J = (J + J.T) h = np.zeros(num_spins) spins = (2*np.random.randint(2, size=(num_spins,)) - 1) def mc_step(spins, J, h, T): current_energy = energy(spins, J, h) for _ in range(spins.shape[0]): i = np.random.randint(spins.shape[0]) dE = energy_diff(i, spins, J, h) if (dE < 0) or (np.random.rand() < np.exp(-dE / T)): current_energy += dE spins[i] *= -1 return spins, current_energy T = 1.0 burn_in = 100 num_samples = 1000 for t in range(burn_in): mc_step(spins, J, h, T) E = np.zeros(num_samples) M = np.zeros(num_samples) for t in range(num_samples): _, e = mc_step(spins, J, h, T) E[t] = e M[t] = np.abs(np.mean(spins)) (np.mean(E), np.std(E)/np.sqrt(num_samples)), (np.mean(M), np.std(M)/np.sqrt(num_samples)) size = num_spins dim = np.arange(2 ** size) space = ((dim[:, None] & (1 << np.arange(size))) > 0)[:, ::-1] space = 2*space.astype(int) - 1 E = energy(space, J, h) M = np.abs(np.mean(space, axis=-1)) logZ = logsumexp(-E / T) probs = np.exp(-E / T - logZ) np.dot(E, probs), np.dot(M, probs) params = pd.read_csv("coefficients.csv", index_col=0)<jupyter_output><empty_output>
permissive
/Project_4_Ising_Annealer/H2_Ising_Annealing.ipynb
HermanniH/CohortProject_2020_Week2
1
<jupyter_start><jupyter_text># Loading mixed data (GTZAN dataset + YouTube data)<jupyter_code>classes = {'reggae': 0, 'classical': 1, 'country': 2, 'disco': 3, 'hiphop': 4, 'jazz': 5, 'metal': 6, 'pop': 7, 'reggae': 8, 'rock': 9} !unzip /content/drive/MyDrive/mel_spec_mix.zip !unzip /content/drive/MyDrive/mel_spec_test.zip genre = os.listdir('/content/mel_spectrogram') os.mkdir('/content/mel_spectrogram/train') os.mkdir('/content/mel_spectrogram/val') random.seed(101) for g in genre: total = list(os.listdir('/content/mel_spectrogram/' + g)) os.mkdir('/content/mel_spectrogram/train/' + g) os.mkdir('/content/mel_spectrogram/val/' + g) random.shuffle(total) train = total[: int(len(total) * 0.85)] val = total[int(len(total) * 0.85) :] for file in train: shutil.copyfile('/content/mel_spectrogram/' + g + '/' + file, '/content/mel_spectrogram/train/' + g + '/' +file) for file in val: shutil.copyfile('/content/mel_spectrogram/' + g + '/' + file, '/content/mel_spectrogram/val/' + g + '/' +file) %rm -rf /content/mel_spectrogram/train/.ipynb_checkpoints %rm -rf /content/mel_spectrogram/val/.ipynb_checkpoints %rm -rf /content/mel_spectrogram_test/.ipynb_checkpoints train_ds = tf.keras.preprocessing.image_dataset_from_directory('/content/mel_spectrogram/train/', batch_size= 32, image_size=(108, 128), ) val_ds = tf.keras.preprocessing.image_dataset_from_directory('/content/mel_spectrogram/val', batch_size= 32, image_size=(108, 128), ) test_ds = tf.keras.preprocessing.image_dataset_from_directory('/content/mel_spectrogram_test', batch_size= 32, image_size=(108, 128), ) plt.figure(figsize=(10, 10)) for images, labels in train_ds.take(1): for i in range(9): ax = plt.subplot(3, 3, i + 1) plt.imshow(images[i].numpy().astype("uint8")) plt.title(genre[labels[i]]) plt.axis("off")<jupyter_output><empty_output><jupyter_text># Convolutional Neural Network Architecture<jupyter_code>model = tf.keras.Sequential([ layers.Conv2D(16, 3, padding='same', activation='relu', input_shape=(108, 128, 3)), layers.MaxPooling2D(), layers.Conv2D(32, 3, padding='same', activation='relu'), layers.MaxPooling2D(), layers.Conv2D(64, 3, padding='same', activation='relu'), layers.MaxPooling2D(), layers.Conv2D(128, 3, padding='same', activation='relu'), layers.MaxPooling2D(), layers.Flatten(), layers.Dropout(0.5), layers.Dense(1024, activation='relu'), layers.Dropout(0.5), layers.Dense(256, activation='relu'), layers.Dense(128, activation='relu'), layers.Dropout(0.5), layers.Dense(32, activation='relu'), layers.Dropout(0.2), layers.Dense(10) ]) model.summary() model.compile(optimizer='adam', loss= tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), metrics=['accuracy']) history = model.fit(train_ds, epochs=40, validation_data= val_ds)<jupyter_output>Epoch 1/40 706/706 [==============================] - 53s 31ms/step - loss: 2.2921 - accuracy: 0.1777 - val_loss: 2.0267 - val_accuracy: 0.2156 Epoch 2/40 706/706 [==============================] - 22s 30ms/step - loss: 1.9776 - accuracy: 0.2778 - val_loss: 1.8370 - val_accuracy: 0.3318 Epoch 3/40 706/706 [==============================] - 22s 32ms/step - loss: 1.8173 - accuracy: 0.3472 - val_loss: 1.6663 - val_accuracy: 0.4115 Epoch 4/40 706/706 [==============================] - 22s 31ms/step - loss: 1.6697 - accuracy: 0.4076 - val_loss: 1.4759 - val_accuracy: 0.4743 Epoch 5/40 706/706 [==============================] - 22s 31ms/step - loss: 1.5765 - accuracy: 0.4514 - val_loss: 1.4054 - val_accuracy: 0.5192 Epoch 6/40 706/706 [==============================] - 22s 31ms/step - loss: 1.4976 - accuracy: 0.4816 - val_loss: 1.3537 - val_accuracy: 0.5294 Epoch 7/40 706/706 [==============================] - 22s 31ms/step - loss: 1.4055 - accuracy: 0.5159 - val_loss: 1.2943 - val_accuracy:[...]<jupyter_text># Accuracy & Loss<jupyter_code>plt.plot(history.history['accuracy']) plt.plot(history.history['val_accuracy']) plt.title('model accuracy') plt.ylabel('accuracy') plt.xlabel('epoch') plt.legend(['train', 'val'], loc='upper left') plt.show() plt.plot(history.history['loss']) plt.plot(history.history['val_loss']) plt.title('model loss') plt.ylabel('loss') plt.xlabel('epoch') plt.legend(['train', 'val'], loc='upper left') plt.show()<jupyter_output><empty_output><jupyter_text># Test on Youtube data<jupyter_code>model.evaluate(test_ds)<jupyter_output>31/31 [==============================] - 1s 17ms/step - loss: 2.2807 - accuracy: 0.4688 <jupyter_text># Saving model<jupyter_code>model.save('/content/drive/MyDrive/ML_Project/models/CNN_Mixed') print('done')<jupyter_output>done
no_license
/sourcecode_dataset/CNN source code/ML_Project_CNN_Mixed.ipynb
tuandung2812/Music-genres-classification
5
<jupyter_start><jupyter_text># 散点图<jupyter_code># 导入包 import numpy as np import matplotlib.pyplot as plt import seaborn as sns import pandas as pd # 创建数据 N = 1000 x = np.random.randn(N) y = np.random.randn(N) # matplotlib作图 plt.scatter(x, y, marker='o') plt.show() plt.scatter(x, y, marker='>') plt.show() plt.scatter(x, y, marker='o') plt.show() # seaborn作图 df = pd.DataFrame({'x':x, 'y':y}) sns.jointplot(x='x', y='y', data= df, kind='scatter') plt.show()<jupyter_output><empty_output>
no_license
/Data-Visualization/ScatterPlot.ipynb
Fengql95/Data-Analysis
1
<jupyter_start><jupyter_text># Tutorial for Small FOV Instruments In this tutorial we combine our skymaps with galaxy catalogs to get a list of galaxies for individual pointings. A note is made that this is only possible with 3D skymaps which are provided for combact binary merger candidate events.We begin by importing the necessary packages as done previously. We will also download the 2MASS Redshift Survey galaxy catalog using VizieR.<jupyter_code># -- For Google Colab #! pip install -q "astropy==3.2.3" "astroquery==0.3.10" "healpy==1.12.9" "matplotlib==3.1.2" "scipy==1.4.1" import healpy as hp # for working with HEALPix files import numpy as np # needed for vector operations from matplotlib import pyplot as plt # plotting skymaps from scipy.stats import norm # probability functions from astropy.utils.data import download_file url = ('https://dcc.ligo.org/public/0146/G1701985/001/LALInference_v2.fits.gz') # This is the publication LALInference localization filename = download_file(url, cache=True)<jupyter_output><empty_output><jupyter_text>We read in the probability, distmu, distsigma, and distnorm.<jupyter_code>prob, distmu, distsigma, distnorm = hp.read_map(filename, field=range(4)) npix = len(prob) nside = hp.npix2nside(npix) npix, nside # Area per pixel in steradians pixarea = hp.nside2pixarea(nside) from astroquery.vizier import Vizier Vizier.ROW_LIMIT = -1 # This gets the complete catalog cat1, = Vizier.get_catalogs('J/ApJS/199/26/table3') # Downloading the 2MRS Galaxy Catalog<jupyter_output><empty_output><jupyter_text>According to Tully(2015), the 2MRS luminosity function is well fit by a Schechter function with a cutoff absolute magnitude of $M_k^* = -23.55$ and a power-law index of $\alpha_K = -1$. We find the maximum absolute magnitude $M_k^{\text{max}}$ for a completeness fraction of 0.5.<jupyter_code>from scipy.special import gammaincinv completeness = 0.5 alpha = -1.0 MK_star = -23.55 MK_max = MK_star + 2.5*np.log10(gammaincinv(alpha + 2, completeness)) MK_max<jupyter_output><empty_output><jupyter_text>Now, we select only galaxies with positive redshifts and absolute magnitudes greater than $M_k^{\text{max}}$.<jupyter_code>from astropy.cosmology import WMAP9 as cosmo from astropy.table import Column import astropy.units as u import astropy.constants as c z = (u.Quantity(cat1['cz'])/c.c).to(u.dimensionless_unscaled) MK = cat1['Ktmag']-cosmo.distmod(z) keep = (z > 0) & (MK < MK_max) cat1 = cat1[keep] z = z[keep]<jupyter_output>/cvmfs/oasis.opensciencegrid.org/ligo/sw/conda/envs/igwn-py37/lib/python3.7/site-packages/astropy/cosmology/core.py:1251: IntegrationWarning: The occurrence of roundoff error is detected, which prevents the requested tolerance from being achieved. The error may be underestimated. args=self._inv_efunc_scalar_args)[0] /cvmfs/oasis.opensciencegrid.org/ligo/sw/conda/envs/igwn-py37/lib/python3.7/site-packages/numpy/lib/function_base.py:2167: RuntimeWarning: invalid value encountered in ? (vectorized) outputs = ufunc(*inputs) /cvmfs/oasis.opensciencegrid.org/ligo/sw/conda/envs/igwn-py37/lib/python3.7/site-packages/astropy/cosmology/core.py:1447: RuntimeWarning: divide by zero encountered in log10 val = 5. * np.log10(abs(self.luminosity_distance(z).value)) + 25.0 /cvmfs/oasis.opensciencegrid.org/ligo/sw/conda/envs/igwn-py37/lib/python3.7/site-packages/astropy/table/column.py:991: RuntimeWarning: invalid value encountered in less result = getattr(super(), op)(other) <jupyter_text>Now, we calculate the luminosity distance and HEALPix index of each galaxy.<jupyter_code>r = cosmo.luminosity_distance(z).to('Mpc').value theta = 0.5*np.pi - cat1['DEJ2000'].to('rad').value phi = cat1['RAJ2000'].to('rad').value ipix = hp.ang2pix(nside, theta, phi)<jupyter_output><empty_output><jupyter_text>We find the probability density per unit volume at the position of each galaxy.<jupyter_code>dp_dV = prob[ipix] * distnorm[ipix] * norm(distmu[ipix], distsigma[ipix]).pdf(r)/pixarea<jupyter_output><empty_output><jupyter_text>Finally, we sort the galaxies by descending probability density and take the top 50.<jupyter_code>top50 = cat1[np.flipud(np.argsort(dp_dV))][:50] top50['RAJ2000', 'DEJ2000', 'Ktmag']<jupyter_output><empty_output><jupyter_text>The coordinates of the first galaxy above are (197.01802, -23.79687). A pointing in this direction would likely have captured the true host galaxy of GW170817 which is (197.45, -23.38).Now, we attempt a similar down-selection but with a different galaxy catalog: the Galaxy List for the Advanced Detector Era.<jupyter_code>catalog_list = Vizier.find_catalogs('GLADE') {k:v.description for k,v in catalog_list.items()} catalogs = Vizier.get_catalogs(catalog_list.keys()) catalogs Vizier.ROW_LIMIT = 50000 # Note, the GLADE catalog is 1,918,147 rows long thus we will get a memory error if we set the row limit to -1 cat2, = Vizier.get_catalogs('VII/275/glade1') # Downloading the GLADE catalog (Galaxy List for the Advanced Detector Era)<jupyter_output><empty_output><jupyter_text>According to Gehrels et al(2016), the GLADE luminosity function is well fit by a Schechter function with a cutoff absolute magnitude of $M_k^* = -20.47$ and a power-law index of $\alpha_K = -1.07$. We find the maximum absolute magnitude $M_k^{\text{max}}$ for a completeness fraction of 0.5.<jupyter_code>completeness = 0.5 alpha = -1.07 MK_star = -20.47 MK_max = MK_star + 2.5*np.log10(gammaincinv(alpha + 2, completeness)) MK_max dist = u.Quantity(cat2['Dist']) # Distance in Mpc z = (u.Quantity(cat2['zph2MPZ'])).to(u.dimensionless_unscaled) MK = cat2['Kmag2']-cosmo.distmod(z) keep = (z > 0) & (MK < MK_max) cat2 = cat2[keep] dist = dist[keep] r = dist.value theta = 0.5*np.pi - cat2['DEJ2000'].to('rad').value phi = cat2['RAJ2000'].to('rad').value ipix = hp.ang2pix(nside, theta, phi) dp_dV = prob[ipix] * distnorm[ipix] * norm(distmu[ipix], distsigma[ipix]).pdf(r)/pixarea top50 = cat2[np.flipud(np.argsort(dp_dV))][:50] top50['RAJ2000', 'DEJ2000', 'Kmag2']<jupyter_output><empty_output>
non_permissive
/skymaps/Tutorial_for_Small_FOV_Instruments.ipynb
battyone/odw-2018
9
<jupyter_start><jupyter_text># Physics 91SI: Lab 3 Part 1 ------<jupyter_code># Don't edit this function def load_sample(): """Return the entire text of Hamlet in a string.""" with open('hamlet.txt') as f: sample = f.read() return sample # Edit this function. "pass" tells Python to do nothing. def parse_sample(sample, words_only=False, sort_list=False): currWord = "" currList = [] for char in sample+" ": c = char if c=="-" or c=="\n": c=" " if c==" ": if currWord != "": currList.append(currWord) currWord = '' continue if words_only: if c.isalpha(): currWord+=(c.lower()) elif c.isnumeric(): currWord+=(c) else: currWord+=(c) if sort_list: currList.sort() return currList parse_sample(load_sample(), True)<jupyter_output><empty_output><jupyter_text>Part 2 ------<jupyter_code>import numpy as np def count_freq(lst): """Return a dict of word frequencies given a list of words. Arguments: -- lst: a list of strings Returns: -- freq_dict: a dictionary of word: frequency pairs. The keys are strings containing the words, and the values are integers indicating how many times each word appears. Example: >>> count_freq(['time', 'after', 'time']) {'time': 2, 'after': 1} """ freq_dict = {} for word in lst: if word not in freq_dict: freq_dict[word]=1 else: freq_dict[word]+=1 return freq_dict def mean(lst): return np.mean(lst) def stdev(lst): return np.std(lst) def median(lst): return np.median(lst)<jupyter_output><empty_output><jupyter_text>Part 3 ------<jupyter_code>def print_stats(lst): ##list of words print("Number of words: ", len(lst)) freq_dict = count_freq(lst) numbers = list(freq_dict.values()) print("Mean word frequency: ", mean(numbers)) print("Standard deviation of word frequencies: ", stdev(numbers)) print("Median word frequency: ", median(numbers)) print_stats(parse_sample(load_sample(), True))<jupyter_output>Number of words: 32195 Mean word frequency: 6.879273504273504 Standard deviation of word frequencies: 38.36281137909605 Median word frequency: 1.0
no_license
/lab3.ipynb
physics91si/lab03-Matias-A
3
<jupyter_start><jupyter_text># INF702 - Text Mining in Social Media /xlsx ( folder) Ontology_PreSNA.xlsx ML_Super.xlsx ML_Unsuper.xlsx Emerging Technologies.xlsx<jupyter_code>import matplotlib.pyplot as plt import numpy as np import seaborn as sns import pandas as pd # Importing Pandas import os outfolder = 'xlsx' if(os.path.exists(outfolder)==False): os.makedirs(outfolder) sns.set() # default set sns.set_style("white") target_label= 'Applications' new_label = 'Application' import matplotlib as mpl mpl.rcParams['font.family'] = 'Calibri' sns.set(font_scale=1.5) sns.set_palette(sns.color_palette("Greens")) fname1 = 'datasets/References1.xlsx' pdf1 = pd.read_excel(fname1) pdf1.head(3) #print(pdf1.info()) #pdf1['Year'] = pdf1['Year'].astype(int) fname2 = 'datasets/References2.xlsx' pdf2 = pd.read_excel(fname2) pdf2.head(3) #pdf2['Year'] = pdf2['Year'].astype(int) #print(pdf2.info()) frames = [pdf1,pdf2] allref=pd.concat(frames) print(len(allref)) allref=allref.reset_index(drop=True) print(len(allref)) #allref.index.droplevel(-1) #print(allref.info()) allref.head(3) #pdf = pdf.replace(r'\\n',' ', regex=True) print(len(allref)) allref = allref.replace('\n',' ', regex=True) allref.head(3) print(len(allref)) print(len(allref)) allref= allref.dropna(axis=0, subset=['Sources']) allref['Sources'].unique() #allref.head(2) print(len(allref)) allref['Year'] = allref['Year'].astype(object) def trim_fraction(text): if '.0' in text: return text[:text.rfind('.0')] return text allref['Year']= allref['Year'].astype(str).apply(trim_fraction) #stat_pd=allref.join(s) #stat_pd.info() #application_df = pd.DataFrame() def getTable1(query, alldf, QueryName, withFile): subdf=alldf.query(query) #print('getTable2 length', len(subdf)) outdf = pd.DataFrame() outdf['Authors'] = subdf['Authors'] outdf['keywords'] = subdf['keywords'] outdf['Applications'] = subdf['Applications'] outdf['Emerging Technologies'] = subdf['Emerging Technologies'] outdf['ML'] = subdf['ML'] outdf['Challenges'] = subdf['Challenges'] outdf['Future Works'] = subdf['Future Works'] outdf['Year'] = subdf['Year'] #adf= application_df.drop_duplicates(keep='first') print('getTable2 length outdf', len(outdf)) writer = pd.ExcelWriter(outfolder+'/'+QueryName+'.xlsx') outdf.to_excel(writer,'Sheet1') writer.save() return outdf #application_df.head(5) def getTable2(query, alldf, QueryName, withFile): subdf=alldf.query(query) #print('getTable2 length', len(subdf)) outdf = pd.DataFrame() outdf['Authors'] = subdf['Authors'] outdf['keywords'] = subdf['keywords'] outdf['Applications'] = subdf['Applications'] outdf['Emerging Technologies'] = subdf['Emerging Technologies'] outdf['ML'] = subdf['ML'] outdf['Challenges'] = subdf['Challenges'] outdf['Future Works'] = subdf['Future Works'] outdf['Year'] = subdf['Year'] #adf= application_df.drop_duplicates(keep='first') print('getTable2 length outdf', len(outdf)) writer = pd.ExcelWriter(outfolder+'/'+ QueryName+'.xlsx') outdf.to_excel(writer,'Sheet1') writer.save() return outdf #application_df.head(5) query = "ML=='Supervised learning'" QueryName = "ML_Super" retd = getTable2(query, allref, QueryName,True) query = "ML=='Unsupervised learning'" QueryName = "ML_Unsuper" retd = getTable2(query, allref, QueryName,True) query = "ML=='Semi-supervised learning'" QueryName = "ML_Semi-super" retd = getTable2(query, allref, QueryName,True) query = "Ontology=='Text Analysis'" QueryName = "OnTology_TA" retd = getTable2(query, allref, QueryName,True) query = "Ontology=='Text Analysis based on SA'" QueryName = "Ontology_TA_SA" retd = getTable2(query, allref, QueryName,True) query = "Ontology=='Prediction based on TA'" QueryName = "Ontology_PreTA" retd = getTable2(query, allref, QueryName,True) query = "Ontology=='Prediction based on SNA'" QueryName = "Ontology_PreSNA" retd = getTable2(query, allref, QueryName,True) query = "Ontology=='Social Network Analysis'" QueryName = "Ontology_SNA" retd = getTable2(query, allref, QueryName,True) df =allref df['index'] = df['Title'].str.find('swarm') #, na = False) #df.head() query = "index !=-1" QueryName = "01Title_swarm" retd = getTable2(query, df, QueryName,True) df =allref df['index'] = df['Title'].str.find('Swarm') #, na = False) #df.head() query = "index !=-1" QueryName = "02Title_Swarm" retd = getTable2(query, df, QueryName,True) df =allref df['ndex'] = df['Title'].str.find('swarm') #, na = False) #df.head() query = "index !=-1" QueryName = "03Keywords_swarm" retd = getTable2(query, df, QueryName,True) df =allref df['index'] = df['Title'].str.find('Swarm') #, na = False) #df.head() query = "index !=-1" QueryName = "04Keywords_Swarm" retd = getTable2(query, df, QueryName,True) df =allref print(len(df)) #query = "Ontology=='Prediction based on SNA'" #subdf=allref.query.substring(query) df['index'] = df['Title'].str.find('swarm') #, na = False) #df.head() query = "index !=-1" QueryName = "Title_swarm" retd = getTable2(query, allref, QueryName,True) #df['Index'] = lambda x: x.find('Super'), df['ML'] #print(len(df)) #print(df['Index']) def getTable3(alldf): #subdf=alldf.query(query) subdf=alldf #print('getTable2 length', len(subdf)) outdf = pd.DataFrame() outdf['Authors'] = subdf['Authors'] outdf['keywords'] = subdf['keywords'] outdf['Applications'] = subdf['Applications'] outdf['Emerging Technologies'] = subdf['Emerging Technologies'] outdf['ML'] = subdf['ML'] outdf['Challenges'] = subdf['Challenges'] outdf['Future Works'] = subdf['Future Works'] outdf['Year'] = subdf['Year'] #adf= application_df.drop_duplicates(keep='first') #print('getTable2 length outdf', len(outdf)) #writer = pd.ExcelWriter(QueryName+'.xlsx') #outdf.to_excel(writer,'Sheet1') #writer.save() return outdf #application_df.head(5) df =allref retd = getTable3(df) df['index'] = df['Emerging Technologies'].isna() #df['index'] #print(df.head(15)) query = "index == False" QueryName = "Emerging Technologies" print(len(df)) retd = getTable2(query, df, QueryName,True) print(len(retd)) #print(retd.head(15)) #print(df['Emerging Technologies'][1]) #df['index'] = df['Emerging Technologies'].isna() #df.head() #query = "index !=-1" #QueryName = "EmergingTechnology" #retd = getTable1(query, allref, QueryName,True) from wordcloud import WordCloud, STOPWORDS, ImageColorGenerator type(retd) #retd = retd.fillna('Text Analysis', regex=True) # Start with one review: text = retd.to_string() #text #text.to_string() #print(type(text))t # Create and generate a word cloud image: wordcloud = WordCloud(background_color='white').generate(text) # Display the generated image: plt.imshow(wordcloud, interpolation='bilinear') plt.axis("off") plt.show()<jupyter_output><empty_output>
no_license
/[Survey]SummaryTable.ipynb
fairhs77/WeeklyTasks
1
<jupyter_start><jupyter_text># Capstone Project This notebook will mainly be used for the Capstone Project in the IBM Data Science Specialization Course on Coursera.<jupyter_code>import pandas as pd import numpy as np print('Hello Capstone Project Course!')<jupyter_output>Hello Capstone Project Course!
no_license
/capstone.ipynb
lmarien/Coursera_Capstone
1
<jupyter_start><jupyter_text># Read in USV data for all 3 Saildrone - caluclate density and wind speed - caluclate distance between successive obs - caluculate total cumulative distance - switch from time to cumulative distance as index - interpolate data onto grid <jupyter_code> ds=[] for iusv in range(3): fname=saildrone_filenames[iusv] ds_usv=xr.open_dataset(fname).isel(trajectory=0).swap_dims({'obs':'time'}) ds_usv.close() ds_usv['wspd']=np.sqrt(ds_usv.UWND_MEAN**2+ds_usv.VWND_MEAN**2) tem=sw.dens0(ds_usv.SAL_SBE37_MEAN,ds_usv.TEMP_SBE37_MEAN) ds_usv['density_mean']=xr.DataArray(tem,dims=('time'),coords={'time':ds_usv.time}) tem=sw.alpha(ds_usv.SAL_SBE37_MEAN,ds_usv.TEMP_SBE37_MEAN,ds_usv.BARO_PRES_MEAN*0) #pressure =0 at surface ds_usv['alpha_ME']=xr.DataArray(tem,dims=('time'),coords={'time':ds_usv.time}) tem=sw.beta(ds_usv.SAL_SBE37_MEAN,ds_usv.TEMP_SBE37_MEAN,ds_usv.BARO_PRES_MEAN*0) #pressure =0 at surface ds_usv['beta_MEAN']=xr.DataArray(tem,dims=('time'),coords={'time':ds_usv.time}) ds_usv['latitude']=ds_usv.latitude.interpolate_na(dim='time') ds_usv['longitude']=ds_usv.longitude.interpolate_na(dim='time') xlat=ds_usv.latitude xlon=ds_usv.longitude dkm2 = abs(np.abs((((xlon[1:].data-xlon[0:-1].data)**2+(xlat[1:].data-xlat[0:-1].data)**2)**.5)*110.567*np.cos(np.pi*xlat[1:].data/180))) dkm2=np.append(dkm2,dkm2[66238]) #add on last point dkm3 = dkm2.cumsum() ds_usv['dist_total']=xr.DataArray(dkm3,dims=('time'),coords={'time':ds_usv.time}) ds_usv['dist_between']=xr.DataArray(dkm2,dims=('time'),coords={'time':ds_usv.time}) if iusv==0: ds = ds_usv else: ds = xr.concat([ds,ds_usv],dim='trajectory')<jupyter_output><empty_output><jupyter_text># Check what the min/max/mean distance travelled between 1 min obs<jupyter_code>for iusv in range(3): print(ds.dist_between[iusv,:].min().data,ds.dist_between[iusv,:].max().data,ds.dist_between[iusv,:].mean().data) #ave distance is 0.08 km = 80 m <jupyter_output>0.0002628831597359855 0.20042159708332527 0.08381687316789584 0.0002626788582655633 0.19503147634695842 0.08436334324573692 8.762569944303782e-05 0.2074201536740167 0.08631113513100724 <jupyter_text># Make an evenly sampled timeseries - Swap the coordinates from time to distance_total - interp along evenly sampled distance total, 80m (0.08km)<jupyter_code>ds_usv = ds.isel(trajectory=0) ds2 = ds_usv.assign_coords(dist_total = ds_usv.dist_total) ds3 = ds2.swap_dims({'time':'dist_total'}) dist_interp = np.arange(ds2.dist_total[0],ds2.dist_total[-1],0.08) ds4 = ds3.interp(dist_total=dist_interp) plt.plot(ds2.time,ds3.density_mean) plt.plot(ds_usv.time,ds_usv.density_mean) <jupyter_output><empty_output><jupyter_text># detrend<jupyter_code>from scipy import signal den = ds4.density_mean.interpolate_na(dim='dist_total') ds4_detrend = signal.detrend(den) #plt.plot(ds4.density_mean) #plt.plot(den) plt.plot(ds4_detrend)<jupyter_output><empty_output><jupyter_text># , smooth using 2km gaussian filter then power density<jupyter_code>import scipy.ndimage ds4_detrend_smooth = scipy.ndimage.filters.gaussian_filter1d(ds4_detrend, sigma=25) plt.plot(ds4_detrend_smooth[5000:7000]) plt.plot(ds4_detrend[5000:7000]) f, Pxx_den = signal.periodogram(ds4_detrend_smooth,1/.080) #fs = sampled at .08km or 80m plt.loglog(f[2:5000], Pxx_den[2:5000]) plt.loglog(f[2:5000], f[2:5000]**(-2.4)/100000) #plt.semilogy(f[2:200], Pxx_den[2:200]) plt.xlabel('frequency [km]') plt.ylabel('PSD [kg/m^3 /km]') length_scale = np.arange(.1,20,.1) xx_in = np.arange(0,.04,.001) xx_in2 = np.arange(0,.04-.001,.001) data = np.ones((len(length_scale),len(xx_in2))) dd=xr.DataArray(data,dims=('length_scale','gradx'),coords={'length_scale':length_scale,'gradx':xx_in2}) ddn=xr.DataArray(data,dims=('length_scale','gradx'),coords={'length_scale':length_scale,'gradx':xx_in2}) ds_usv = ds.isel(trajectory=0) ds2 = ds_usv.assign_coords(dist_total = ds_usv.dist_total) ds3 = ds2.swap_dims({'time':'dist_total'}) for ilen2,len2 in enumerate(length_scale): dist_interp = np.arange(ds2.dist_total[0],ds2.dist_total[-1],len2) ds4 = ds3.interp(dist_total=dist_interp) den_grad = np.abs(np.gradient(ds4.density_mean)) result,xx = np.histogram(den_grad,bins=xx_in) dd[ilen2,:]=result dd[ilen2,:]=result/sum(result) print(len(length_scale),len(xx_in)) fig = plt.figure(figsize=(10,10)) tem=ddn tem = tem.where(tem>.04) plt.pcolor(length_scale,xx_in2,tem.T,vmin=0,vmax=.2,cmap='hot') plt.colorbar()<jupyter_output>199 40
permissive
/atomic/Calculate_density_plot.ipynb
agonmer/Saildrone
5
<jupyter_start><jupyter_text># Growing Decision Trees via ID3/C4.5 & Greedy Learning A from-scratch implementation of a growing decision trees ML algorithm for quick classification. We use entropy / 'information gain' as the metric for generating the decision rules. This notebook is a step-by-step walkthrough, with motivating graphs. ### General Structure of notebook: * Introduce entropy function associated with the information gain metric that will be used to implement growing decision trees. * Preliminary decision tree implementation * Final decision tree implementation * *Note:* Right now the decision tree implementation assumes a dataframe structure with: * the (desired) class label column having index 0 (first column) * an additional final (last indexed) column of aggregate counts There is nothing special about these requirements, and the code may eventually be rewritten to be a little more flexible (allow you to indicate the class label column at your disgression and let the counts column be located anywhere). As is discussed later, the class label column allows the dataset to be compressed and makes computations more efficient. Furthermore, I include a walkthrough of the relatively straightforward process of transforming a raw dataset into a compressed form with aggregate counts, and wrap it in a function which makes it straightforwards to apply. Also note this implementation is not optimized to run efficiently on large data. ### Example An application of growing decision trees on a sample dataset. I've included some sample datasets to test the implementation on.<jupyter_code>### Example Bank Dataset import pandas as pd ### Entering data as columns - pandas reads them as tuples though so they will have to be transformed credit_ranking_data = ["Excellent","Excellent","Excellent","Excellent","Good","Good","Good","Good","Fair","Fair","Fair","Fair"] age_data = ["< 30","< 30","> 60","> 60","30 - 60","30 - 60","< 30","< 30","> 60","30 - 60","< 30","< 30"] gender_data = ["M","F","M","F","M","F","M","F","M","F","M","F"] year_income_data = ["60k - 100k","> 100k","> 100k","60k - 100k","< 60k","60k - 100k","> 100k","60k - 100k","> 100k","60k - 100k","60k - 100k","< 60k"] counts_data = [16,4,4,16,5,15,5,15,2,2,18,18] ex_df = pd.DataFrame.from_records(data = [ credit_ranking_data, age_data, gender_data, year_income_data, counts_data]) ### Transpose to put in proper form ex_df = ex_df.T ### Add column labels cols = ["credit-ranking","age","gender","year-income","count"] ex_df.columns = cols ex_df # just checking<jupyter_output><empty_output><jupyter_text>The bank-dataset table above is a useful illustration of the kind of formatting the algorithm I will write expects: * The desired class-label column is the first column (this can be adjusted to be more flexible in the future) * The last column is a count column (this can be constructed without too much difficulty by using some aggregation functions on arbitrary unstructured datasets, and compresses the dataset to make the algorithm run more efficiently) In the following, I walk through construction of the information purity heuristic which we use to grow the decision tree. ## Information Purity Measures<jupyter_code>import math # term from info function I def p_logp(p): # scope --- p is proportion from 0 to 1 if (p < 10**(-10)) | (p > 1 - 10**(-10)) : return 0 else: return (-1) * p * math.log2(p) <jupyter_output><empty_output><jupyter_text>We can use the "p_logp" function to compute the statistical information entropy (physical measure) that occurs from observing a proportion $p$ of observations:<jupyter_code>## EXAMPLE BASIC ENTROPY FUNCTION granularity = 100 domain = [x/granularity for x in range(granularity+1)] entropy_range = [ p_logp(p) + p_logp(1-p) for p in domain ] # print(entropy_range) import matplotlib.pyplot as plt plt.scatter(domain,entropy_range) plt.ylabel('entropy = -[plog(p) + (1-p)log(1-p)]') plt.show() # information measure def info_measure(vector): ### NOTE: vector should be of counts, so all terms in it ≥ 0!! ADD CODE TO CHECK THIS LATER # if vector == [0]*len(vector): return 0 if sum(vector) == 0: return 0 return sum( [ p_logp( s/sum(vector) ) for s in vector ] ) <jupyter_output><empty_output><jupyter_text>## The Information measure $I(\mathbf{x}) = - \sum \left(\frac{x_i}{ \sum x_i } \right) \log_2 \left(\frac{x_i}{ \sum x_i } \right)$ generalizes this entropy measure to proportions drawn from higher-dimension source - ie, more than the two states corresponding to $p$ and $1-p$ (the interpretation of $\mathbf{x}$ is as a list of counts for the different states $i$). Thinking back to the basic entropy function we introduced, which came from our '$p \log_2 p$' function, if $p$ is the proportion $k/n$ then the basic entropy function was $I(\mathbf{x})$ where $\mathbf{x} = (k,n-k)$. In general, this measures the amount of information required to classify a tuple $\mathbf{x}$ to $m$ classes (if the input $\mathbf{x}$ is an $m$-vector). ### An Example: Consider: plotting the information measure over an input of datapoints $(x,y)$ uniformly sampled from the part of a 2D circle with $x,y \geq 0$ (since $x,y$ both represent counts (and the point $(x,y)$ then a configuration of counts in the system / data), then it makes sense although it is not necessary mathematically, to restrict our attention to a specific 'positive domain' subset of the circle. <jupyter_code>## EXAMPLE of the INFORMATION Measure on the input of a 2D circle from mpl_toolkits.mplot3d import Axes3D import matplotlib.pyplot as plt import random def circleplot(granularity,color=[random.random() for i in range(3)]): # granularity = 10000 domain = [[x/granularity, (1-(x/granularity)**2)**0.5] for x in range(granularity+1)] entropy_range = [ info_measure(x) for x in domain ] # print(entropy_range) fig = plt.figure() ax = fig.add_subplot(1,1,1, projection='3d') circ_points = [ x + [z] for x,z in zip(domain,entropy_range) ] xs = [x[0] for x in circ_points] ys = [x[1] for x in circ_points] zs = [x[2] for x in circ_points] ### different rectangular parametrizations of a 2d-circle ### we plot entropy associated with a particular configuration of points ax.scatter( xs,ys,zs , c=color) ### Can plot along all points on a circle - but don't have interpretations as negative counts # ax.scatter( xs,[-y for y in ys],zs , c=color ) # ax.scatter( [-x for x in xs],ys,zs , c=color) # ax.scatter( [-x for x in xs],[-y for y in ys],zs , c=color) ax.set_xlabel('X') ax.set_ylabel('Y') ax.set_zlabel('Z') # plt.scatter([x[0] for x in domain],entropy_range) # plt.scatter([x[1] for x in domain],entropy_range) # plt.ylabel('some numbers') plt.show() circleplot(10000, 'goldenrod') # shortcut code<jupyter_output><empty_output><jupyter_text>## Another example: Plot the information metric for classifying a mesh of 2D points $(x,y)$ over a rectangular region $x,y \in [0,1]$ (aka $[0,1] \times [0,1]$).<jupyter_code>from mpl_toolkits.mplot3d import Axes3D import matplotlib.pyplot as plt granularity = 25 domain = [[x/granularity, y/granularity] for x in range(granularity+1) for y in range(granularity+1)] entropy_range = [ info_measure(x) for x in domain ] # print(entropy_range) fig = plt.figure() ax = fig.add_subplot(1,1,1, projection='3d') rect_points = [ x + [z] for x,z in zip(domain,entropy_range) ] xs = [x[0] for x in rect_points] ys = [x[1] for x in rect_points] zs = [x[2] for x in rect_points] ### we plot entropy associated with a particular configuration of points ax.scatter( xs,ys,zs , c="goldenrod") # ### plot circle for comparison # ##### # granularity = 100 # domain = [[x/granularity, (1-(x/granularity)**2)**0.5] for x in range(granularity+1)] # entropy_range = [ info_measure(x) for x in domain ] # circ_points = [ x + [z] for x,z in zip(domain,entropy_range) ] # xs = [x[0] for x in circ_points] # ys = [x[1] for x in circ_points] # zs = [x[2] for x in circ_points] # ### different rectangular parametrizations of a 2d-circle # ### we plot entropy associated with a particular configuration of points # ax.scatter( xs,ys,zs , c='blue') # # ax.scatter( xs,[-y for y in ys],zs , c='blue' ) # # ax.scatter( [-x for x in xs],ys,zs , c='blue') # # ax.scatter( [-x for x in xs],[-y for y in ys],zs , c='blue') # ##### ax.set_xlabel('X') ax.set_ylabel('Y') ax.set_zlabel('Z') # plt.scatter([x[0] for x in domain],entropy_range) # plt.scatter([x[1] for x in domain],entropy_range) # plt.ylabel('some numbers') plt.show()<jupyter_output><empty_output><jupyter_text>Comparison of Plots: [TBD]## Entropy Function Next, we want to use the information measure to formally define an entropy function for attributes of a dataset, which is based off a weighted sum of the aggregate information measures for a given configuration of data: If the attribute $A$ with values $\{ a_1 , \dots , a_\nu \}$ partitions the dataset into subsets $\{S_1, \dots , S_\nu\}$, and $S_j$ contains $s_{ij}$ samples of a given class class $C_i$, then the entropy associated with $A$ is $$ E ( A ) = \sum_{j=1} ^\nu \frac{ s_{1j} + \cdots + s_{mj} }{\sum_{ij} s_{ij}} I(s_{1j} , \dots , s_{mj}) $$ Some observations –– from intuition (and by studying its defining equation) we can see that $I ( \mathbf{x} )$ tends to be maximized along the line $x_1 = x_2 = \cdots = x_m$, and hence so does the entropy function. As the interpretation of these $x_i$ was as counts for different states $i$, this means roughly speaking that we maximize entropy when the counts are similar and minimize it when they are very different. The metric we will use to grow the decision tree is information purity, or information gain, which increases as entropy decreases. So, we will maximize this when we split over branches where the counts tend to be most dissimilar. This all makes sense too, as if the counts are very dissimilar, than intuitively it is easier to predict a class label based on the counts while if they are similar, it is more difficult to decide what an appropriate class label should be.<jupyter_code># entropy cost associated with an attribute # ASSUME: last column of input dataset is 'count' associated with particular outcome # ASSUME: first column is class label def entropy(dataset, attribute, debug_flag=False): label = dataset.iloc[:,0] # 1st column class label total_counts = sum(dataset['count']) # REPLACE 'count' eventually total_entropy = 0 for value in set(dataset[attribute]): if debug_flag: print(attribute, value) #debugging sub_data = dataset.loc[dataset[attribute] == value] if debug_flag: print(sub_data) #debugging label_counts = [] for label_value in set(label): # pick out rows from sub data with desired label value sub_sub_data = sub_data.loc[sub_data.iloc[:,0] == label_value] label_counts.append(sum(sub_sub_data['count'])) # REPLACE 'count' eventually # if debug_flag: print(label_counts) #debugging if debug_flag: print(str(sum(label_counts)) + '/' + str(total_counts) + ' * I(' + str(label_counts) + ')') #debugging total_entropy += (sum(label_counts)/total_counts) * info_measure(label_counts) if debug_flag: print('*'*50) return total_entropy ### ENTROPY on example dataset print(ex_df) print() print('*' * 50) print() ## Possible attributes to compute entropy on: ## Remove class label and count columns from consideration possible_attributes = list(ex_df.columns) possible_attributes = [possible_attributes[i] for i in range(1,len(possible_attributes)-1)] print((possible_attributes[0], 'entropy: ' + str(entropy(ex_df, possible_attributes[0], debug_flag = False)))) # Define Information Gain Criteria # equivalent to minimizing entropy but has a nice interpretation def info_gain(dataset, attribute, debug_flag = False): #get label counts label = dataset.iloc[:,0] # assume it's first column for now label_counts = [] for label_value in set(label): # pick out rows from dataset with desired label value sub_data = dataset.loc[dataset.iloc[:,0] == label_value] label_counts.append(sum(sub_data['count'])) # REPLACE 'count' eventually if debug_flag: print('I(...):', label_counts) #debugging print('*'*50) #debugging I = info_measure(label_counts) return I - entropy(dataset, attribute, debug_flag) info_gain(ex_df,possible_attributes[2],debug_flag=True) # debug flag displays info_measure coefficients for attribute in possible_attributes: print(( attribute, 'entropy: ' + str(entropy(ex_df,attribute)), 'gain: ' + str(info_gain(ex_df,attribute)) )) # information gained by branching on attribute A<jupyter_output>('age', 'entropy: 1.1265494549903445', 'gain: 0.45841304573081154') ('gender', 'entropy: 1.5421864522230475', 'gain: 0.042776048498108565') ('year-income', 'entropy: 1.3836778130106993', 'gain: 0.20128468771045682') <jupyter_text>## Growing Decision Trees: Now that we have a properly defined information gain metric for a given attribute of a dataset, we're prepared to use this to implement growing a decision tree such that it maximizes local information gain on each attribute, at each step. Note: We should be careful to avoid overgrowing the tree (various methods not yet implemented) as well as to look into methods to post/pre prune a tree in order to avoid over-fitting the decision tree model on the data. For example, we could implement a $k$-fold cross validation method to grow $k$ different trees on train/test folds and compare the overall accuracy obtained, and try boosting to improve on the accuracy of the model. ### Printing decision trees Here is a first pass at a growing-decision-tree algorithm, that prints generated decision rules as well as debugging info. It will automatically grow decisions to full length (use every attribute) even when not necessary. This overgrowing will be optimized for the proper algorithm.<jupyter_code>#### GROWING DECISION TREES: def print_decision_tree(dataset, condition_text='', debug_flag = False): # condition_text used to pass conditions down the desicion tree (for printing) if debug_flag: print(dataset) # for debugging columns = list(dataset.columns) # label = columns[0] # counts = columns[len(columns)] # get possible attributes for branching possible_attributes = columns possible_attributes = [possible_attributes[i] for i in range(1,len(possible_attributes)-1)] # remove first entry (class label) and last (count) if len(possible_attributes) > 0: ### can still keep branching # maximize info gain on possible attributes info_gains = [info_gain(dataset, attribute) for attribute in possible_attributes] # as first index for label wasn't included in info_gains, we do +1 to get the correct index branching_index = info_gains.index(max(info_gains)) + 1 new_condition = 'branch on ' + str(columns[branching_index]) # the new attribute to branch on # print( new_condition ) # this info is printed as 'rule' below branching_attribute = dataset.iloc[:,branching_index] for attribute_value in set(branching_attribute): sub_dataset = dataset.loc[ dataset.iloc[:,branching_index] == attribute_value] sub_dataset = sub_dataset.drop(columns[branching_index], axis=1) rule = condition_text + ' ' + new_condition + ': ' + str(attribute_value) print(rule) # print(attribute_value) # iterate print_decision_tree( sub_dataset, rule, debug_flag ) else: # no more attributes to split on, so take majority vote to decide what the correct label is votes = [] for outcome in set(dataset.iloc[:,0]): sub_dataset = dataset.loc[dataset.iloc[:,0] == outcome] votes.append( [outcome, sum(sub_dataset['count'])] ) # REPLACE 'count' eventually temp = [x[1] for x in votes] decision_index = temp.index(max(temp)) decision = votes[decision_index][0] print( 'final decision: ' + str(decision)) print('*'*50) return # returns null-type # Example ex_df print_decision_tree(ex_df) <jupyter_output> branch on age: 30 - 60 branch on age: 30 - 60 branch on gender: F branch on age: 30 - 60 branch on gender: F branch on year-income: 60k - 100k final decision: Good ************************************************** branch on age: 30 - 60 branch on gender: M branch on age: 30 - 60 branch on gender: M branch on year-income: < 60k final decision: Good ************************************************** branch on age: < 30 branch on age: < 30 branch on year-income: < 60k branch on age: < 30 branch on year-income: < 60k branch on gender: F final decision: Fair ************************************************** branch on age: < 30 branch on year-income: > 100k branch on age: < 30 branch on year-income: > 100k branch on gender: F final decision: Excellent ************************************************** branch on age: < 30 branch on year-income: > 100k branch on gender: M final decision: Good ************************************************** branch on age: < 30 branch on year-in[...]<jupyter_text>### Evolving the function slightly After the first pass, we're ready to amend some flaws. Right now I have set up a decision tree builder using recursion. It outputs a two item list –– the first is a list of 'verbose' decision branches (as strings) and the second is a condensed form of the decision tree as lists. Probably eventually I will rewrite the second form to either remove it entirely, or make its output more amenable for direct analysis. <jupyter_code>### Initial Pass # still need to alter algorithm to prevent 'overgrowing' the tree when unnecessary def decision_tree(dataset, condition_text='', decision_rules_human_readable=[], decision_rules_as_list=[]): columns = list(dataset.columns) possible_attributes = columns possible_attributes = [possible_attributes[i] for i in range(1,len(possible_attributes)-1)] # remove first entry (class label) and last (count) ## len(list(set(dataset.iloc[:,0]))) == '#' of out omes in class label boolean = not ( len(set(dataset.iloc[:,0])) == 1 ) # ie there are more than 1 class labels in the data still if (len(possible_attributes) > 0 & boolean): ### can keep branching # maximize info gain on possible attributes info_gains = [info_gain(dataset, attribute) for attribute in possible_attributes] branching_index = info_gains.index(max(info_gains)) + 1 # as first index for label wasn't included in info_gains new_condition = str(columns[branching_index]) branching_attribute = dataset.iloc[:,branching_index] for attribute_value in set(branching_attribute): sub_dataset = dataset.loc[ dataset.iloc[:,branching_index] == attribute_value] sub_dataset = sub_dataset.drop(columns[branching_index], axis=1) rule = condition_text + '%.%' + new_condition + '%.%' + str(attribute_value) decision_rules_as_list += [[ columns[branching_index], attribute_value ]] # If it's a 'complete decision rule' add it to the list if len(possible_attributes) == 1: decision_rules_human_readable.append( [rule], ) decision_tree( sub_dataset, rule, decision_rules_human_readable, decision_rules_as_list ) else: # no more attributes to split on, assign label via majority vote votes = [] for outcome in set(dataset.iloc[:,0]): sub_dataset = dataset.loc[dataset.iloc[:,0] == outcome] votes.append( [outcome, sum(sub_dataset['count'])] ) # REPLACE 'count' eventually temp = [x[1] for x in votes] decision_index = temp.index(max(temp)) decision = votes[decision_index][0] decision_rules_human_readable += [' final decision:' + str(decision)] decision_rules_as_list += [[columns[0], decision]] return [decision_rules_human_readable, decision_rules_as_list]<jupyter_output><empty_output><jupyter_text>Let's test building decision trees on our example dataset:<jupyter_code>trees = decision_tree(ex_df) trees[0]<jupyter_output><empty_output><jupyter_text>### Breakdown and Accuracy I wrote a quick tool that uses the 'verbose' stem of the decision tree builder to compute the accuracy of the decision rule numerically, as well as in fraction form (so that we can easily skim through effect size). I may rewrite it slightly in the future in order to either: * perhaps use the list form output of the decision tree * directly leave in effect size so that output can be sorted for highest accuracy *and* effect size simultaneously<jupyter_code>def accuracy(dataset, label): correct_counts = sum( dataset.loc[ dataset.iloc[:, 0] == label ]['count'] ) total_counts = sum( dataset['count'] ) # if total_counts==0: print(dataset) # print(correct_counts, total_counts) acc = correct_counts / total_counts return (acc, str(correct_counts) + '/' + str(total_counts)) def analyze(dataset, decision_rules_list, importance_filter=False): result = [] columns = list(dataset.columns) # print('length of decision list is ' + str(len(decision_rules_list))) for i in range( int(len(decision_rules_list) / 2) ): parsed = [x.split('%.%') for x in decision_rules_list[2*i]] # get attributes from decision rule branch # get subdataset satisfying parsed rules conditions = parsed[0][1:] #throw away 0th term = a space subdataset = dataset for j in range( int(len(conditions)/2) ): # -1 to prevent inex to go 1 iteration too many (and pass in empty arg) index = columns.index(conditions[2*j]) # print(i,j) # print(conditions) # print(columns[index], conditions[2*j+1]) subdataset = subdataset.loc[ subdataset.iloc[:,index] == conditions[2*j + 1]] # print('8'*20) decision = decision_rules_list[2*i+1].split(':')[1] # print(decision) # print(subdataset) result.append(accuracy(subdataset, decision)) i = i + 2 # new rule starts every two entries trees_and_accuracy = [ [decision_rules_list[2*i] + [decision_rules_list[2*i+1]] , result[i]] for i in range(len(result)) ] # sorts by accuracy, decreasing. set importance threshold at say .1% of database size? if importance_filter: temp = sorted(trees_and_accuracy, key=lambda x: x[1], reverse = True) final_info = [x for x in temp if (int(x[1][1][0]) > 15) ] else: final_info = sorted(trees_and_accuracy, key=lambda x: x[1], reverse = True) return final_info ### Outputs list of decision branches + accuracy analyze(ex_df, trees[0])<jupyter_output><empty_output><jupyter_text># A Few More Examples Let's see the algorithm in practice on a few more datasets and extend it beyond the 'toy' dataset ex_df we've used so far. ## 2. Working with a raw dataset from scratch I first generate a (skewed) dataset using the label-pools from ex_df, and put it into the standard form my decision tree algorithm utilizes before building decision trees and analyzing the results:<jupyter_code>### Entering data as columns - pandas reads them as tuples though so they will have to be transformed credit_ranking_data = ["Excellent","Good","Fair"] age_data = ["> 60","30 - 60","< 30"] gender_data = ["M","F"] year_income_data = ["> 100k","60k - 100k","< 60k"] # counts_data # I will build the counts aggregate_attributes = [credit_ranking_data, age_data, gender_data, year_income_data] ### simulate_data: import random n = 10000 data = [] for i in range(n): # to make it more interesting / realistic, I am going to skew it slightly if i % 15 == 0: data.append(["Excellent", random.choice(age_data[0:2]), "F", random.choice(year_income_data[1:2])]) else: data.append( [random.choice(x) for x in aggregate_attributes] ) df = pd.DataFrame.from_records(data) # ### Add column labels cols = ["credit-ranking","age","gender","year-income"]#,"count"] df.columns = cols df.head() # just checking <jupyter_output><empty_output><jupyter_text>### Compressing the raw dataset I'm going to build the counts column and shrink the dataset in size. This helps make the algorithm scale since we don't have to access the entire dataset.<jupyter_code>compressed_df = df[ [not x for x in df.duplicated()] ] # following is good for sorting but runs into a copy-setting warning, so I've just left it out for the moment # compressed_df["credit-ranking"] = pd.Categorical(compressed_df['credit-ranking'], ["Excellent","Good","Fair"]) # sorts by variable, in order compressed_df = compressed_df.sort_values(["credit-ranking", 'age', 'gender']) print('length:', len(compressed_df)) compressed_df.head() ## Add counts: # reminder about cols = columns counts = [] for index, row in compressed_df.iterrows(): #count sub_df = df for i in range(len(cols)): sub_df = sub_df.loc[ df[cols[i]] == row[i] ] count = sum( sub_df.duplicated() ) counts.append(count + 1) # +1 since undercounting duplicates by 1 (removes the original comparison) compressed_df['count'] = counts print('total counts:' , sum(compressed_df['count'])) # had better sum to the original 10000 compressed_df[0:15] # first 15 terms tree = decision_tree(compressed_df) # tree[0] # stores the verbose decision trees analyze(compressed_df, tree[0])[0:10] <jupyter_output><empty_output><jupyter_text>Some interesting (and relatively weak) results --- unsurprising considering how much of the data was randomly generated. Lets wrap the compression procedure above in a function we can reuse:<jupyter_code>def structure_data(dataframe): ### # Assumes a pandas dataframe ### Compress dataframe compressed_df = dataframe[ [not x for x in dataframe.duplicated()] ] columns = list(dataframe.columns) # sorts by variable, in order compressed_df = compressed_df.sort_values(columns) # # debug # print('length:', len(compressed_df)) # compressed_df.head() ### Add counts: counts = [] for index, row in compressed_df.iterrows(): #count sub_dataframe = dataframe for i in range(len(columns)): sub_dataframe = sub_dataframe.loc[ dataframe[columns[i]] == row[i] ] count = sum( sub_dataframe.duplicated() ) counts.append(count + 1) # +1 since undercounting duplicates by 1 (removes the original comparison) compressed_df['count'] = counts # debug print('total counts:' , sum(compressed_df['count']), 'dataset length:' , len(dataframe)) return compressed_df <jupyter_output><empty_output><jupyter_text>## 3. Working with a (slightly) larger dataset The Food Mart dataset contains approximately 2000 transactions containing subsets of 100 different items - a simplified version of a small market transaction database TDB. Each row is a transaction, and a 1 in the tuple indicates the item was purchased, a 0 indicates it wasn't. One feature to notice in this dataset is the sparseness of the TDB --- most transactions will only contain a few of the 100 various products.<jupyter_code># Example 3 Food_Mart = pd.read_excel('FoodMart.xls') print(Food_Mart.columns) Food_Mart.head()<jupyter_output>Index(['Acetominifen', 'Anchovies', 'Aspirin', 'Auto Magazines', 'Bagels', 'Batteries', 'Beer', 'Bologna', 'Candles', 'Canned Fruit', 'Canned Vegetables', 'Cereal', 'Cheese', 'Chips', 'Chocolate', 'Chocolate Candy', 'Clams', 'Cleaners', 'Coffee', 'Cold Remedies', 'Computer Magazines', 'Conditioner', 'Cookies', 'Cooking Oil', 'Cottage Cheese', 'Crackers', 'Deli Meats', 'Deli Salads', 'Deodorizers', 'Dips', 'Donuts', 'Dried Fruit', 'Dried Meat', 'Eggs', 'Fashion Magazines', 'Flavored Drinks', 'French Fries', 'Fresh Chicken', 'Fresh Fish', 'Frozen Chicken', 'Frozen Vegetables', 'Gum', 'Hamburger', 'Hard Candy', 'Home Magazines', 'Hot Dogs', 'Ibuprofen', 'Ice Cream', 'Jam', 'Jelly', 'Juice', 'Lightbulbs', 'Maps', 'Milk', 'Mouthwash', 'Muffins', 'Nasal Sprays', 'Nuts', 'Oysters', 'Pancake Mix', 'Pancakes', 'Paper Dishes', 'Paper Wipes', 'Pasta', 'Peanut Butter', 'Personal Hygiene', 'Pizza', 'Plastic U[...]<jupyter_text>It is difficult to run the unoptimized compressor on the full dataset, so instead let us restrict our attention to a subset: <jupyter_code>columns_subset = list(Food_Mart.columns)[35:65] small_df = pd.DataFrame(data = [ Food_Mart[x] for x in columns_subset ]).T # [ # Food_Mart['Cooking Oil'], Food_Mart['Eggs'], # Food_Mart['Frozen Chicken'], Food_Mart['Frozen Vegetables'], # Food_Mart['Pancake Mix'], Food_Mart['Sauces'], # Food_Mart['Pot Scrubbers'], Food_Mart['Soup'], # Food_Mart['Sour Cream'], Food_Mart['Spices'], # Food_Mart['Sugar']]).T # T for transpose df = structure_data(small_df) df.head()<jupyter_output>total counts: 2127 dataset length: 2127 <jupyter_text>Let's suppose we also want to predict whether French Fries are purchased –– we can reorder the columns as well:<jupyter_code>### Reorder old_cols = df.columns.tolist() index_salads = old_cols.index('French Fries') cols = [old_cols[index_salads]] + [ old_cols[i] for i in range(len(old_cols)) if i != index_salads ] fm_df = df[cols] fm_df.head() fm_tree = decision_tree(fm_df,'',[],[]) # have to overwrite stored values from previous use fm_tree[0][0:5] analyze(fm_df, fm_tree[0])[0:5] <jupyter_output>/anaconda3/lib/python3.6/site-packages/pandas/core/ops.py:798: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison result = getattr(x, name)(y)
no_license
/Growing Decision Trees Implementation.ipynb
Kodyak/Growing-Decision-Trees
16
<jupyter_start><jupyter_text>## INF1608 - Análise Numérica - 2017.1 ## Departamento de Informática - PUC-Rio ## Prof. Hélio Lopes - [email protected] ## http://www.inf.puc-rio.br/~lopes # Exercícios resolvidos R1) Faça uma função que verique se uma matriz A de tamanho nxn é estritamente diagonal dominante: Definição: Uma matriz A (nxn) é estritamente diagonal dominante se para toda linha i vale: $$|A_{i,i}| > \sum_{i=1, i\ne j}^n |A_{i,j}|$$ R2) Implemente o método de Jacobi para a solução de um sistema de equações lineares. Defina um critério de parada para o seu método, e explique-o. https://pt.wikipedia.org/wiki/M%C3%A9todo_de_Jacobi R3) Implemente o método de Gauss-Seidel para a solução de um sistema de equações lineares. Defina um critério de parada para o seu método, e explique-o. https://pt.wikipedia.org/wiki/M%C3%A9todo_de_Gauss-Seidel ## Lista 4 1) Implemente o método de Gradiente Conjugado para a solução de um sistema de equações lineares. Defina um critério de parada para o seu método, e explique-o. https://en.wikipedia.org/wiki/Conjugate_gradient_method 2) Teste muito bem todos esses algoritmos!## SoluçãoR1) Faça uma função que verique se uma matriz A de tamanho nxn é estritamente diagonal dominante: Definição: Uma matriz A (nxn) é estritamente diagonal dominante se para toda linha i vale: $$|A_{i,i}| > \sum_{i=1, i\ne j}^n |A_{i,j}|$$ <jupyter_code>import numpy as np import math def isdiagonaldominat(A): n = len(A) for i in range(n): soma = 0.0 for j in range(n): if (i != j): a = A[i,j] soma += math.fabs(a) if (math.fabs(A[i,i]) <= soma): return False return True A = np.matrix([[-2.0,1.0,0.0],[1.0,-2.0,1.0],[0.0,1.0,-2.0]]) print(A) print(isdiagonaldominat(A)) <jupyter_output>[[-2. 1. 0.] [ 1. -2. 1.] [ 0. 1. -2.]] False <jupyter_text>R2) Implemente o método de Jacobi para a solução de um sistema de equações lineares. Defina um critério de parada para o seu método, e explique-o. https://pt.wikipedia.org/wiki/M%C3%A9todo_de_Jacobi<jupyter_code>import numpy as np def jacobi(A,b,x0,N,TOL): nlin = len(A); #num. de linhas x = np.zeros(nlin); #cria x #iteracoes k = 1; while (k <= N): #iteracao de Jacobi for i in range(nlin): x[i] = 0.0; for j in range(i): x[i] = x[i] - A[i,j]*x0[j] for j in range(i+1,nlin): x[i] = x[i] - A[i,j]*x0[j] x[i] = (x[i] + b[i])/A[i,i] #criterio de parada if (np.linalg.norm(x-x0) < TOL): return x,k #prepara nova iteracao k = k+1; x0 = np.copy(x) return x0,k A = np.matrix([[2.0,1.0],[5.0,7.0]]) b = np.array([11.0,13.0]) guess = np.array([1.0,1.0]) sol,n = jacobi(A,b,guess,100,0.000001) print "A:" print(A) print "b:" print(b) print "x:" print(sol) print "||Ax-b||:" print(np.linalg.norm(A.dot(sol)-b)) print "Numero de iteracoes:" print(n) <jupyter_output>A: [[ 2. 1.] [ 5. 7.]] b: [ 11. 13.] x: [ 7.1111107 -3.22222137] ||Ax-b||: 3.92347242466e-06 Numero de iteracoes: 31 <jupyter_text>R3) Implemente o método de Gauss-Seidel para a solução de um sistema de equações lineares. Defina um critério de parada para o seu método, e explique-o. https://pt.wikipedia.org/wiki/M%C3%A9todo_de_Gauss-Seidel<jupyter_code>import numpy as np def gaussseidel(A,b,x0,N,TOL): nlin = len(A); #num. de linhas x = np.zeros(nlin); #cria x #iteracoes k = 1; while (k <= N): #iteracao de Jacobi for i in range(nlin): s = 0.0; for j in range(i): s = s + A[i,j]*x[j] for j in range(i+1,nlin): s = s + A[i,j]*x0[j] x[i] = (b[i] - s)/A[i,i] #criterio de parada if (np.linalg.norm(x-x0) < TOL): return x,k #prepara nova iteracao k = k+1; x0 = np.copy(x) return x0,k A = np.matrix([[2.0,1.0],[5.0,7.0]]) b = np.array([11.0,13.0]) guess = np.array([1.0,1.0]) sol,n = gaussseidel(A,b,guess,100,0.000001) print "A:" print(A) print "b:" print(b) print "x:" print(sol) print "||Ax-b||:" print(np.linalg.norm(A.dot(sol)-b)) print "Numero de iteracoes:" print(n) <jupyter_output>A: [[ 2. 1.] [ 5. 7.]] b: [ 11. 13.] x: [ 7.1111107 -3.22222193] ||Ax-b||: 5.3245767262e-07 Numero de iteracoes: 16 <jupyter_text>1) Implemente o método de Gradiente Conjugado para a solução de um sistema de equações lineares. Defina um critério de parada para o seu método, e explique-o. https://en.wikipedia.org/wiki/Conjugate_gradient_method 2) Teste muito bem todos esses algoritmos!<jupyter_code>import numpy as np def conjugateGradient(A,b,x0,N,TOL): n = len(A) x = np.zeros(n) k = 0 r0 = np.copy(b - A.dot(x0)) p = np.copy(r0) while(k < N): alpha = r0.dot(r0)/p.dot(A.dot(p)) x = np.copy(x0 + alpha * p) r = np.copy(r0 - alpha * A.dot(p)) if (np.linalg.norm(r) < TOL): return x, k p = np.copy(r + p * r.dot(r) / r0.dot(r0)) # nova iteração k = k + 1 x0 = np.copy(x) r0 = np.copy(r) return x0, k # GERAR UMA MATRIZ SIMÉTRICA POSITIVA DEFINIDA DIAGONAL DOMINANTE PARA FINS DE TESTE D = np.array([[13.,0.,0.,0.],[0.,2.,0.,0.],[0.,0.,5.,0.],[0.,0.,0.,11.]]) # Matriz diagal de autovalores positivos P = np.array([[1.,1.,1.,1.],[1.,1.,-1.,-1.],[-1.,1.,-1.,1.],[1.,-1.,-1.,1.]]) # Matriz de colunas ortogonais P = P/np.sqrt(4.)# Normalizar colunas para gerar matriz ortogonal Pt = P.transpose() Pinv = np.linalg.inv(P) print ("Pt - Pinv: (Deve ser 0 para P ser ortogonal)\n" + str(Pt-Pinv)) # Teste se P é ortogonal A = Pt.dot(D.dot(P))# A = P^(-1)DP #A = np.array([[1577.0,3460.0,753.0],[3460.0,37052.0,4278.0],[753.0,4278.0,594.0]]) b = np.array([21.0,13.0,12.0,11.0]) chute = np.array([1.0,1.0,1.0,1.0]) #np.linalg.cholesky(A) # <- Só para garantir que a matriz é positiva definida sol,n = conjugateGradient(A,b,chute,100,0.000001) print "A: (Matriz gerada em bloco de código acima simérica, positiva definida)" print(A) print "b:" print(b) print "x:" print(sol) print "||Ax-b||:" print(np.linalg.norm(A.dot(sol)-b)) print "Numero de iteracoes:" print(n)<jupyter_output>Pt - Pinv: (Deve ser 0 para P ser ortogonal) [[ 0. 0. 0. 0.] [ 0. 0. 0. 0.] [ 0. 0. 0. 0.] [ 0. 0. 0. 0.]] A: (Matriz gerada em bloco de código acima simérica, positiva definida) [[ 7.75 -0.25 1.25 4.25] [-0.25 7.75 4.25 1.25] [ 1.25 4.25 7.75 -0.25] [ 4.25 1.25 -0.25 7.75]] b: [ 21. 13. 12. 11.] x: [ 3.08024476 1.86206294 0.01206294 -0.56975524] ||Ax-b||: 0.0 Numero de iteracoes: 3
no_license
/Exercicios lista metodos iterativos- Rafael.ipynb
rafarubim/analise-numerica
4
<jupyter_start><jupyter_text># Sample Jupyter Notebook to call COVID data provider <jupyter_code>func_url = 'http://localhost:7071/api/covidata' import pandas as pd from ipywidgets import Image import requests df = pd.read_csv(func_url+"?country=US") df.head() Image(value=requests.get(func_url+"?country=US&output=plot").content)<jupyter_output><empty_output>
no_license
/notebooks/sample.ipynb
shwars/e2e_covid
1
<jupyter_start><jupyter_text>## Image pyramids Take a look at how downsampling with image pyramids works. First, we'll read in an image then construct and display a few layers of an image pyramid.<jupyter_code>import numpy as np import matplotlib.pyplot as plt import cv2 %matplotlib inline # Read in the image image = cv2.imread('images/rainbow_flag.jpg') # Change color to RGB (from BGR) image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) plt.imshow(image) level_1 = cv2.pyrDown(image) level_2 = cv2.pyrDown(level_1) level_3 = cv2.pyrDown(level_2) # Display the images f, (ax1,ax2,ax3,ax4) = plt.subplots(1, 4, figsize=(20,10)) ax1.set_title('original') ax1.imshow(image) ax2.imshow(level_1) ax2.set_xlim([0, image.shape[1]]) ax2.set_ylim([0, image.shape[0]]) ax3.imshow(level_2) ax3.set_xlim([0, image.shape[1]]) ax3.set_ylim([0, image.shape[0]]) ax4.imshow(level_3) ax4.set_xlim([0, image.shape[1]]) ax4.set_ylim([0, image.shape[0]]) <jupyter_output><empty_output>
permissive
/1_4_Feature_Vectors/1. Image Pyramids.ipynb
svedagiriml/CVND_Exercises
1
<jupyter_start><jupyter_text># Lesson 1: Introduction to Deep Learning with PyTorch - [@AlfredoCanziani](https://twitter.com/alfredocanziani) - [@GokuMohandas](https://twitter.com/GokuMohandas)### Create the data<jupyter_code>import numpy as np import tensorflow as tf from tensorflow import keras from tensorflow.keras import layers from sklearn.model_selection import train_test_split import matplotlib.pyplot as plt import random import math seed=12345 random.seed(seed) # torch.manual_seed(seed) tf.random.set_seed(seed) N = 1000 # num_samples_per_class D = 2 # dimensions C = 3 # num_classes H = 100 # num_hidden_units X = [] y = [] for i in range(C): index = 0 r = np.linspace(0, 1, N) t = np.linspace( i * 2 * math.pi / C, (i + 2) * 2 * math.pi / C, N ) + np.random.normal(N) * 0.1 for ix in range(N * i, N * (i + 1)): X.append(r[index] * np.array([math.sin(t[index]), math.cos(t[index])])) y.append(i) index += 1 X = tf.stack(X) y = tf.stack(y) print("SHAPES:") print("-------------------") print("X:", tuple(X.shape)) print("y:", tuple(y.shape)) def plot_data(X, y, d=.0, auto=False): """ Plot the data. """ plt.clf() plt.scatter(X[:, 0], X[:, 1], c=y, s=20, cmap=plt.cm.Spectral) plt.axis('square') plt.axis((-1.1, 1.1, -1.1, 1.1)) if auto is True: plt.axis('equal') # plt.savefig('spiral{:.2f}.png'.format(d)) # Create the data plot_data(X.numpy(), y.numpy()) def plot_model(X, y, model, e=.0, auto=False): """ Plot the model from torch weights. """ X = X.numpy() y = y.numpy(), w1 = model.layers[0].get_weights()[0] b1 = model.layers[0].get_weights()[1] w2 = model.layers[1].get_weights()[0] b2 = model.layers[1].get_weights()[1] h = 0.01 x_min, x_max = (-1.1, 1.1) y_min, y_max = (-1.1, 1.1) if auto is True: x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1 y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1 xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h)) Z = np.dot(np.maximum(0, np.dot(np.c_[xx.ravel(), yy.ravel()], w1) + b1), w2) + b2 Z = np.argmax(Z, axis=1) Z = Z.reshape(xx.shape) fig = plt.figure() plt.contourf(xx, yy, Z, cmap=plt.cm.Spectral, alpha=0.3) plt.scatter(X[:, 0], X[:, 1], c=y, s=40, cmap=plt.cm.Spectral) plt.axis((-1.1, 1.1, -1.1, 1.1)) plt.axis('square') if auto is True: plt.axis((xx.min(), xx.max(), yy.min(), yy.max())) # plt.savefig('train{:03.2f}.png'.format(e))<jupyter_output><empty_output><jupyter_text>### Linear model<jupyter_code>learning_rate = 1e-3 lambda_l2 = 1e-5 def linear_model(D_in, H, D_out): model = keras.Sequential([ layers.Dense(H, activation='linear', input_shape=D_in), layers.Dense(D_out, activation='linear'), ]) optimizer = tf.keras.optimizers.SGD(learning_rate=learning_rate) model.compile(loss='mse', optimizer=optimizer) return model model_1 = linear_model(X.shape,H,C) model_1.fit(X, y, epochs=100) model_1.summary() # Plot trained model plot_model(X, y, model)<jupyter_output><empty_output><jupyter_text>### Two-layered network<jupyter_code>learning_rate = 1e-3 lambda_l2 = 1e-5 def two_layer_network(D_in, H, D_out): model = keras.Sequential([ layers.Dense(H, activation='relu', input_shape=D_in), layers.Dense(D_out), ]) optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate) model.compile(loss='binary_crossentropy', optimizer=optimizer) return model model_2 = two_layer_network(X.shape, H, C) model_2.fit(X,y, epochs = 100) model_2.summary() # Plot trained model plot_model(X, y, model_2)<jupyter_output><empty_output><jupyter_text>### Proper training procedure#### Create datasets<jupyter_code>split_ratio = 0.8 # train-test split num_epochs = 100 batch_size = 64 log_every = 25 training = random.sample(range(X.shape[0]), int(X.shape[0] * split_ratio)) testing = [i for i in range(X.shape[0]) if not i in training] X_train = tf.convert_to_tensor(X.numpy()[training,:]) X_test = tf.convert_to_tensor(X.numpy()[testing,:]) y_train = tf.convert_to_tensor(y.numpy()[training]) y_test = tf.convert_to_tensor(y.numpy()[testing]) X_train.shape, y_train.shape<jupyter_output><empty_output><jupyter_text>#### Training<jupyter_code>learning_rate = 1e-3 lambda_l2 = 1e-5 dropout_p = 0.1 decay_rate = 0.9999 max_grad_norm = 5.0 def customized_network(D_in, H, D_out): model = keras.Sequential([ layers.Dense(H, activation='relu', input_shape=D_in), layers.Dense(D_out,activation='softmax'), layers.Dropout(dropout_p), ]) optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate) model.compile(loss='binary_crossentropy', optimizer=optimizer) return model model_3 = customized_network(X_train.shape, H, C) model_3.fit(X_train,y_train, epochs = 100) model_3.summary() # Plot trained model plot_model(X, y, model_3)<jupyter_output><empty_output>
no_license
/HW1/Pytorch and Tensorflow/lesson1_Tensorflow.ipynb
tunghoangt/DeepNeuralNet
5
<jupyter_start><jupyter_text># Preparando dados no HDFS<jupyter_code>!ls !hdfs dfs -ls / !hdfs dfs -mkdir /semantix !hdfs dfs -ls / !ls -lha !hdfs dfs -put NASA* /semantix !hdfs dfs -ls /semantix<jupyter_output>Found 2 items -rw-r--r-- 1 root hdfs 167813770 2019-08-17 20:45 /semantix/NASA_access_log_Aug95 -rw-r--r-- 1 root hdfs 205242368 2019-08-17 20:45 /semantix/NASA_access_log_Jul95 <jupyter_text># Carregando dados no spark<jupyter_code>df1 = spark.read.text("/semantix/NASA_access_log_Aug95") df1.show(10) df2 = spark.read.text("/semantix/NASA_access_log_Jul95") df2.show(10)<jupyter_output>+--------------------+ | value| +--------------------+ |199.72.81.55 - - ...| |unicomp6.unicomp....| |199.120.110.21 - ...| |burger.letters.co...| |199.120.110.21 - ...| |burger.letters.co...| |burger.letters.co...| |205.212.115.106 -...| |d104.aa.net - - [...| |129.94.144.152 - ...| +--------------------+ only showing top 10 rows <jupyter_text>## Verificando carregamento<jupyter_code>df1.count() df2.count()<jupyter_output><empty_output><jupyter_text>## Juntando os arquivos<jupyter_code>df = df1.union(df2) df.count()<jupyter_output><empty_output><jupyter_text># Tratamento dos dados e organização do dataframe<jupyter_code>df_table = df.select( F.regexp_extract('value', r'^([^\s]+\s)', 1).alias('HOST'), F.regexp_extract('value', r'^.*\[(\d\d/\w{3}/\d{4}:\d{2}:\d{2}:\d{2} -\d{4})]', 1).alias('TIMESTAMP'), F.regexp_extract('value', r'^.*"\w+\s+([^\s]+)\s+HTTP.*"', 1).alias('REQUISITION'), F.regexp_extract('value', r'^.*"\s+([^\s]+)', 1).alias('STATUS'), F.regexp_extract('value', r'^.*\s+(\d+)$', 1).alias('SIZE') ) df_table.show() df_table.count()<jupyter_output><empty_output><jupyter_text>## Tratamento de data e tipagem<jupyter_code>df_table = df_table\ .where(F.trim(df_table.TIMESTAMP) != "")\ .select(df_table.HOST, F.when(df_table.TIMESTAMP.like("%Aug%"),\ F.from_unixtime(F.unix_timestamp(\ F.regexp_replace(df_table.TIMESTAMP, "Aug", "08"), 'dd/MM/yyyy')) )\ .otherwise(F.from_unixtime(F.unix_timestamp( F.regexp_replace(df_table.TIMESTAMP, "Jul", "07"), 'dd/MM/yyyy')))\ .cast("date")\ .alias("DATE"), df_table.REQUISITION, df_table.STATUS.cast('integer'), df_table.SIZE.cast('integer') ) df_table.show()<jupyter_output>+--------------------+----------+--------------------+------+-----+ | HOST| DATE| REQUISITION|STATUS| SIZE| +--------------------+----------+--------------------+------+-----+ | in24.inetnebr.com |1995-08-01|/shuttle/missions...| 200| 1839| | uplherc.upl.com |1995-08-01| /| 304| 0| | uplherc.upl.com |1995-08-01|/images/ksclogo-m...| 304| 0| | uplherc.upl.com |1995-08-01|/images/MOSAIC-lo...| 304| 0| | uplherc.upl.com |1995-08-01|/images/USA-logos...| 304| 0| |ix-esc-ca2-07.ix....|1995-08-01|/images/launch-lo...| 200| 1713| | uplherc.upl.com |1995-08-01|/images/WORLD-log...| 304| 0| |slppp6.intermind....|1995-08-01|/history/skylab/s...| 200| 1687| |piweba4y.prodigy....|1995-08-01|/images/launchmed...| 200|11853| |slppp6.intermind....|1995-08-01|/history/skylab/s...| 200| 9202| |slppp6.intermind....|1995-08-01|/images/ksclogosm...| 200| 3635| |ix-esc-ca2-07.ix....|1995-08-01|/history/apollo[...]<jupyter_text>## Questões### 1. Número de hosts únicos.<jupyter_code>df_table.select(df_table.HOST).distinct().count()<jupyter_output><empty_output><jupyter_text>### 2. O total de erros 404.<jupyter_code>df_table.where(df_table.STATUS == 404).count()<jupyter_output><empty_output><jupyter_text>### 3. Os 5 URLs que mais causaram erro 404.<jupyter_code>df_table.select(df_table.HOST,df_table.STATUS)\ .where(df_table.STATUS == 404)\ .groupBy(df_table.HOST)\ .count()\ .withColumnRenamed("count", "COUNT_404")\ .orderBy("COUNT_404", ascending = False)\ .select("HOST", "COUNT_404")\ .show(5)<jupyter_output>+--------------------+---------+ | HOST|COUNT_404| +--------------------+---------+ |hoohoo.ncsa.uiuc....| 251| |piweba3y.prodigy....| 157| |jbiagioni.npt.nuw...| 132| |piweba1y.prodigy....| 114| |www-d4.proxy.aol....| 91| +--------------------+---------+ only showing top 5 rows <jupyter_text>### 4. Quantidade de erros 404 por dia.<jupyter_code>df_table.select(df_table.DATE, df_table.STATUS)\ .where(df_table.STATUS == 404)\ .groupBy(df_table.DATE)\ .count()\ .withColumnRenamed("count", "COUNT_404")\ .orderBy("DATE")\ .show(30)<jupyter_output>+----------+---------+ | DATE|COUNT_404| +----------+---------+ |1995-07-01| 316| |1995-07-02| 291| |1995-07-03| 474| |1995-07-04| 359| |1995-07-05| 497| |1995-07-06| 640| |1995-07-07| 570| |1995-07-08| 302| |1995-07-09| 348| |1995-07-10| 398| |1995-07-11| 471| |1995-07-12| 471| |1995-07-13| 532| |1995-07-14| 413| |1995-07-15| 254| |1995-07-16| 257| |1995-07-17| 406| |1995-07-18| 465| |1995-07-19| 639| |1995-07-20| 428| |1995-07-21| 334| |1995-07-22| 192| |1995-07-23| 233| |1995-07-24| 328| |1995-07-25| 461| |1995-07-26| 336| |1995-07-27| 336| |1995-07-28| 94| |1995-08-01| 243| |1995-08-03| 304| +----------+---------+ only showing top 30 rows <jupyter_text>### 5. O total de bytes retornados.<jupyter_code>df_table.select(F.sum(df_table.SIZE).alias("TOTAL")).show() gb=1073741824 size_table = df_table.select(F.sum(df_table.SIZE).alias("TOTAL")) size_table.select("TOTAL").withColumn('TOTAL_GB', (size_table.TOTAL/gb).cast('integer')).show()<jupyter_output>+-----------+--------+ | TOTAL|TOTAL_GB| +-----------+--------+ |65524314915| 61| +-----------+--------+
no_license
/ProcessoSemantix.ipynb
atworkdante/semantix
11
<jupyter_start><jupyter_text># Tensor Transformations<jupyter_code>from __future__ import print_function import torch import numpy as np from datetime import date date.today() torch.__version__ np.__version__<jupyter_output><empty_output><jupyter_text>NOTE on notation * _x, _y, _z, ...: NumPy 0-d or 1-d arrays * _X, _Y, _Z, ...: NumPy 2-d or higer dimensional arrays * x, y, z, ...: 0-d or 1-d tensors * X, Y, Z, ...: 2-d or higher dimensional tensors## CastingQ1. Convert a torch Tensor `X` to numpy array `_X`.<jupyter_code>X = torch.randn(3,2,1) _X = X.numpy() print(type(_X))<jupyter_output><class 'numpy.ndarray'> <jupyter_text>Q2. Convert a torch Tensor X to a python list Y.<jupyter_code>X = torch.randn(3,2,1) Y = X.tolist() print(type(Y))<jupyter_output><class 'list'> <jupyter_text>Q3. Convert the data type of X to float32.<jupyter_code>X = torch.IntTensor(3, 2).fill_(0) print("X=>", X.type()) Y = X.float() print("Y=>", Y.type())<jupyter_output>X=> torch.IntTensor Y=> torch.FloatTensor <jupyter_text>Q4. Convert the data type of X to int32.<jupyter_code>X = torch.LongTensor(3, 2).fill_(0) print("X=>", X.type()) Y = ... print("Y=>", Y.type())<jupyter_output>X=> torch.LongTensor Y=> torch.IntTensor <jupyter_text>## Size and ResizeQ5. Get the size and type of X.<jupyter_code>X = torch.randn(3, 2, 1) y = X.shape tp = X.type() print("size=", y) print("type=", tp) <jupyter_output>size= torch.Size([3, 2, 1]) type= torch.FloatTensor <jupyter_text>Q6. Get the total number of elements in X.<jupyter_code>X = torch.randn(3, 2, 1) y = X.numel() print(y) <jupyter_output>6 <jupyter_text>Q7. Get the number of dimensions of X.<jupyter_code>X = torch.randn(3, 2, 1) y = X.ndimension() print(y) <jupyter_output>3 <jupyter_text>Q8. Resize X to (10, 3).<jupyter_code>X = torch.ones(5, 6) Y = X.resize_(10, 3) print(Y.shape)<jupyter_output>torch.Size([10, 3]) <jupyter_text>Q9. Remove all the dimensions of size 1 in X.<jupyter_code>X = torch.randn(10, 10, 1, 1) Y = X.squeeze() print(Y.size()) print(X.size())<jupyter_output>torch.Size([10, 10]) torch.Size([10, 10, 1, 1]) <jupyter_text>Q10. Remove only the third dimension in X.<jupyter_code>X = torch.randn(10, 10, 1, 1) Y = X.squeeze(2) # ??torch.squeez2 print(Y.size())<jupyter_output>torch.Size([10, 10, 1]) <jupyter_text>Q11. Add a dimension of 1 at the end of X.<jupyter_code>X = torch.ones(10, 10, 3) Y = X.unsqueeze(-1) print(Y.size())<jupyter_output>torch.Size([10, 10, 3, 1]) <jupyter_text>## Indexing, Slicing, Joining, Mutating OpsQ12. Return the columns of indices y.<jupyter_code>X = torch.arange(1, 22).resize_(3, 7) print(X) y = torch.LongTensor([0, 2, 4, 5]) print(X.index_select(1, y)) ??torch.index_select<jupyter_output> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 [torch.FloatTensor of size 3x7] 1 3 5 6 8 10 12 13 15 17 19 20 [torch.FloatTensor of size 3x4] <jupyter_text>Q13. Apply `mask` to `x`.<jupyter_code>x = torch.arange(1, 11) print("x=", x) mask = x.gt(5) print("masked=", x.masked_select(mask)) <jupyter_output>x= 1 2 3 4 5 6 7 8 9 10 [torch.FloatTensor of size 10] masked= 6 7 8 9 10 [torch.FloatTensor of size 5] <jupyter_text>Q14. Split X into 5 same-sized tensors along the second dimension.<jupyter_code>X = torch.arange(1, 11).resize_(2, 5) print("X=", X) Ys = torch.chunk(X, 5, 1) print("Chunks=", Ys)<jupyter_output>X= 1 2 3 4 5 6 7 8 9 10 [torch.FloatTensor of size 2x5] Chunks= ( 1 6 [torch.FloatTensor of size 2x1] , 2 7 [torch.FloatTensor of size 2x1] , 3 8 [torch.FloatTensor of size 2x1] , 4 9 [torch.FloatTensor of size 2x1] , 5 10 [torch.FloatTensor of size 2x1] ) <jupyter_text>Q16. Remove the second dimension of X and return all resulting slices.<jupyter_code>X = torch.arange(1, 11).resize_(2, 5) print("X=", X) ys = torch.unbind(X, 1) print("Sliced tensors=", ys)<jupyter_output>X= 1 2 3 4 5 6 7 8 9 10 [torch.FloatTensor of size 2x5] Sliced tensors= ( 1 6 [torch.FloatTensor of size 2] , 2 7 [torch.FloatTensor of size 2] , 3 8 [torch.FloatTensor of size 2] , 4 9 [torch.FloatTensor of size 2] , 5 10 [torch.FloatTensor of size 2] ) <jupyter_text>Q17. Stack x, y, and z vertically.<jupyter_code>x = torch.Tensor([1, 4]) y = torch.Tensor([2, 5]) z = torch.Tensor([3, 6]) O = torch.stack((x, y, z)) print(O) <jupyter_output> 1 4 2 5 3 6 [torch.FloatTensor of size 3x2] <jupyter_text>Q18. Repeat X 1 and 2 times along each dimension .<jupyter_code>X = torch.arange(1, 7).resize_(2, 3) print("X=", X) print(X.dim()) Y = X.repeat(1, 2) print("Repeated=", Y) <jupyter_output>X= 1 2 3 4 5 6 [torch.FloatTensor of size 2x3] 2 Repeated= 1 2 3 1 2 3 4 5 6 4 5 6 [torch.FloatTensor of size 2x6] <jupyter_text>Q19. Concatenate X and Y along the first dimension.<jupyter_code>X = torch.Tensor([[1, 2, 3], [4, 5, 6]]) Y = torch.Tensor([[7, 8, 9], [10, 11, 12]]) Z = torch.cat((X, Y), dim=0) print(Z)<jupyter_output> 1 2 3 4 5 6 7 8 9 10 11 12 [torch.FloatTensor of size 4x3] <jupyter_text>Q20. Swap the first and second dimension of X.<jupyter_code>X = torch.randn(3, 2) print("X=", X) Y = X.t() print("Transposed=", Y)<jupyter_output>X= 0.0214 -0.0313 -0.7030 -0.4384 -2.0496 -0.1647 [torch.FloatTensor of size 3x2] Transposed= 0.0214 -0.7030 -2.0496 -0.0313 -0.4384 -0.1647 [torch.FloatTensor of size 2x3] <jupyter_text>Q21. Permute X's dimensions such that the new tensor has shape (3, 1, 2).<jupyter_code>X = torch.ones(1, 2, 3) Y = X.permute(2, 0, 1) print(Y.size()) <jupyter_output>torch.Size([3, 1, 2]) <jupyter_text>## SortingQ22. Sort X along the second dimension.<jupyter_code>X = torch.Tensor([[1,4],[3,1]]) print("X=", X) sorted_tensor, sorted_indices = ... print("sorted tensor=", sorted_tensor) print("sorted indices=", sorted_indices) <jupyter_output>X= 1 4 3 1 [torch.FloatTensor of size 2x2] sorted tensor= 1 4 1 3 [torch.FloatTensor of size 2x2] sorted indices= 0 1 1 0 [torch.LongTensor of size 2x2] <jupyter_text>## CountingQ23. Get the indices of all nonzero elements in X.<jupyter_code>X = torch.Tensor([[0,1,7,0,0],[3,0,0,2,19]]) y = ... print(y) <jupyter_output> 0 1 0 2 1 0 1 3 1 4 [torch.LongTensor of size 5x2]
permissive
/0-libraries/pytorch-exercises/Chapter1_Tensors/1-tensor-transformations.ipynb
Christomesh/deep-learning-implementations
23
<jupyter_start><jupyter_text>##### Copyright 2019 The TensorFlow Authors.<jupyter_code>#@title Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # https://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License.<jupyter_output><empty_output><jupyter_text># TensorFlow 2 quickstart for beginners View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook This short introduction uses [Keras](https://www.tensorflow.org/guide/keras/overview) to: 1. Build a neural network that classifies images. 2. Train this neural network. 3. And, finally, evaluate the accuracy of the model.This is a [Google Colaboratory](https://colab.research.google.com/notebooks/welcome.ipynb) notebook file. Python programs are run directly in the browser—a great way to learn and use TensorFlow. To follow this tutorial, run the notebook in Google Colab by clicking the button at the top of this page. 1. In Colab, connect to a Python runtime: At the top-right of the menu bar, select *CONNECT*. 2. Run all the notebook code cells: Select *Runtime* > *Run all*.Download and install TensorFlow 2. Import TensorFlow into your program: Note: Upgrade `pip` to install the TensorFlow 2 package. See the [install guide](https://www.tensorflow.org/install) for details.<jupyter_code>import tensorflow as tf<jupyter_output><empty_output><jupyter_text>Load and prepare the [MNIST dataset](http://yann.lecun.com/exdb/mnist/). Convert the samples from integers to floating-point numbers:<jupyter_code>mnist = tf.keras.datasets.mnist (x_train, y_train), (x_test, y_test) = mnist.load_data() x_train, x_test = x_train / 255.0, x_test / 255.0<jupyter_output>Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz 8192/11490434 [..............................] - ETA: 0s<jupyter_text>Build the `tf.keras.Sequential` model by stacking layers. Choose an optimizer and loss function for training:<jupyter_code>model = tf.keras.models.Sequential([ tf.keras.layers.Flatten(input_shape=(28, 28)), tf.keras.layers.Dense(128, activation='relu'), tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(10) ])<jupyter_output><empty_output><jupyter_text>For each example the model returns a vector of "[logits](https://developers.google.com/machine-learning/glossary#logits)" or "[log-odds](https://developers.google.com/machine-learning/glossary#log-odds)" scores, one for each class.<jupyter_code>predictions = model(x_train[:1]).numpy() predictions<jupyter_output><empty_output><jupyter_text>The `tf.nn.softmax` function converts these logits to "probabilities" for each class: <jupyter_code>tf.nn.softmax(predictions).numpy()<jupyter_output><empty_output><jupyter_text>Note: It is possible to bake this `tf.nn.softmax` in as the activation function for the last layer of the network. While this can make the model output more directly interpretable, this approach is discouraged as it's impossible to provide an exact and numerically stable loss calculation for all models when using a softmax output. The `losses.SparseCategoricalCrossentropy` loss takes a vector of logits and a `True` index and returns a scalar loss for each example.<jupyter_code>loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)<jupyter_output><empty_output><jupyter_text>This loss is equal to the negative log probability of the true class: It is zero if the model is sure of the correct class. This untrained model gives probabilities close to random (1/10 for each class), so the initial loss should be close to `-tf.log(1/10) ~= 2.3`.<jupyter_code>loss_fn(y_train[:1], predictions).numpy() model.compile(optimizer='adam', loss=loss_fn, metrics=['accuracy'])<jupyter_output><empty_output><jupyter_text>The `Model.fit` method adjusts the model parameters to minimize the loss: <jupyter_code>model.fit(x_train, y_train, epochs=5)<jupyter_output>Epoch 1/5 <jupyter_text>The `Model.evaluate` method checks the models performance, usually on a "[Validation-set](https://developers.google.com/machine-learning/glossary#validation-set)" or "[Test-set](https://developers.google.com/machine-learning/glossary#test-set)".<jupyter_code>model.evaluate(x_test, y_test, verbose=2)<jupyter_output>313/313 - 0s - loss: 0.0825 - accuracy: 0.9753 <jupyter_text>The image classifier is now trained to ~98% accuracy on this dataset. To learn more, read the [TensorFlow tutorials](https://www.tensorflow.org/tutorials/).If you want your model to return a probability, you can wrap the trained model, and attach the softmax to it:<jupyter_code>probability_model = tf.keras.Sequential([ model, tf.keras.layers.Softmax() ]) probability_model(x_test[:5])<jupyter_output><empty_output>
no_license
/TF-Iris-rasp3b/beginner.ipynb
Neelanjan-Goswami/AWS-Sagemaker_-_AWS-IoT-Analytics
11
<jupyter_start><jupyter_text>### BasicsThe basics of Python Programming* #### Printing Stuff<jupyter_code>print("Hello, World!")<jupyter_output>Hello, World! <jupyter_text>* #### String substitution<jupyter_code>print("Hello, {0}, have a good {1}.".format("Saurav", "day")) print("Hello, {name}, have a good {time}.".format(time="day", name="Saurav")) name = "Saurav" print("Hello ", name)<jupyter_output>Hello Saurav <jupyter_text>* #### Taking Input<jupyter_code>name = input("Enter your name: ") age = int(input("Enter your age: "))<jupyter_output>Enter your age: 23 <jupyter_text>* #### Operators<jupyter_code>ans = 1 + 2 - 3 * 4 / 5 // 6 % 7 ** 8 print(ans) 2 <= 3 and 4 > 5 or 6 != 7<jupyter_output><empty_output><jupyter_text>* #### Data TypesPython int data types can represent numbers as big as your virtual memory. However, representing floats may result in prescision errors. Boolean values True and False are equal to numerical values 1 and 0. Strings can be any sequence of characters between two quotes, single and double both included. They can be sliced, index, concatanated or repeated.<jupyter_code>name = 'Saurav' len(name) name[2] name[-1] name + name name * 3 name[0:3] name[1:3] name[2:] name[:4]<jupyter_output><empty_output><jupyter_text>* #### Lists, Tuples, Dicts<jupyter_code>list = [1, 2, 3, 4] len(list) list[2] list[1:2] tuple = ("abc", "xyz") len(tuple) dict = {"name": "Saurav", "age": 24} len(dict) dict.keys()<jupyter_output><empty_output><jupyter_text>* #### Conditionals<jupyter_code>a = True if a == True: print("True") elif a == False: print("False") else: print("Unknown")<jupyter_output>True <jupyter_text>* #### Loops<jupyter_code>a = 0 while a <5: print(a) a+=1 for i in range(5): print(i) name = "Saurav" for i, j in enumerate(name): print(i, j)<jupyter_output>0 S 1 a 2 u 3 r 4 a 5 v <jupyter_text>### Functions<jupyter_code>def addNumbers(x, y): return x + y addNumbers(2, 2)<jupyter_output><empty_output><jupyter_text>A function can also take default arguments.<jupyter_code>def addNumbers(x=2, y=3): return x +y addNumbers()<jupyter_output><empty_output><jupyter_text>### Classes And Objects<jupyter_code>class Calculator(object): def __init__(self, x, y): self.x = x self.y = y def add(self): return self.x + self.y def subtract(self): return self.x - self.y c = Calculator(2, 3) print(c.add()) print(c.subtract())<jupyter_output>5 -1
no_license
/CorePython/00_Basics.ipynb
uniquerockrz/learning-python
11
<jupyter_start><jupyter_text>## 1. Import and observe dataset We all love watching movies! There are some movies we like, some we don't. Most people have a preference for movies of a similar genre. Some of us love watching action movies, while some of us like watching horror. Some of us like watching movies that have ninjas in them, while some of us like watching superheroes. Movies within a genre often share common base parameters. Consider the following two movies: Both movies, 2001: A Space Odyssey and Close Encounters of the Third Kind, are movies based on aliens coming to Earth. I've seen both, and they indeed share many similarities. We could conclude that both of these fall into the same genre of movies based on intuition, but that's no fun in a data science context. In this notebook, we will quantify the similarity of movies based on their plot summaries available on IMDb and Wikipedia, then separate them into groups, also known as clusters. We'll create a dendrogram to represent how closely the movies are related to each other. Let's start by importing the dataset and observing the data provided.<jupyter_code># Import modules import numpy as np import pandas as pd import nltk # Set seed for reproducibility np.random.seed(5) # Read in IMDb and Wikipedia movie data (both in same file) movies_df = pd.read_csv("datasets/movies.csv") print("Number of movies loaded: %s " % (len(movies_df))) # Display the data movies_df.head()<jupyter_output>Number of movies loaded: 100 <jupyter_text>## 2. Combine Wikipedia and IMDb plot summaries The dataset we imported currently contains two columns titled wiki_plot and imdb_plot. They are the plot found for the movies on Wikipedia and IMDb, respectively. The text in the two columns is similar, however, they are often written in different tones and thus provide context on a movie in a different manner of linguistic expression. Further, sometimes the text in one column may mention a feature of the plot that is not present in the other column. For example, consider the following plot extracts from The Godfather: Wikipedia: "On the day of his only daughter's wedding, Vito Corleone" IMDb: "In late summer 1945, guests are gathered for the wedding reception of Don Vito Corleone's daughter Connie" While the Wikipedia plot only mentions it is the day of the daughter's wedding, the IMDb plot also mentions the year of the scene and the name of the daughter. Let's combine both the columns to avoid the overheads in computation associated with extra columns to process.<jupyter_code># Combine wiki_plot and imdb_plot into a single column movies_df["plot"] = movies_df["wiki_plot"].astype(str) + "\n" + \ movies_df["imdb_plot"].astype(str) # Inspect the new DataFrame movies_df.head()<jupyter_output><empty_output><jupyter_text>## 3. Tokenization Tokenization is the process by which we break down articles into individual sentences or words, as needed. Besides the tokenization method provided by NLTK, we might have to perform additional filtration to remove tokens which are entirely numeric values or punctuation. While a program may fail to build context from "While waiting at a bus stop in 1981" (Forrest Gump), because this string would not match in any dictionary, it is possible to build context from the words "while", "waiting" or "bus" because they are present in the English dictionary. Let us perform tokenization on a small extract from The Godfather.<jupyter_code># Tokenize a paragraph into sentences and store in sent_tokenized sent_tokenized = [sent for sent in nltk.sent_tokenize(""" Today (May 19, 2016) is his only daughter's wedding. Vito Corleone is the Godfather. """)] # Word Tokenize first sentence from sent_tokenized, save as words_tokenized words_tokenized = [word for word in nltk.word_tokenize(sent_tokenized[0])] # Remove tokens that do not contain any letters from words_tokenized import re filtered = [word for word in words_tokenized if re.search('[a-zA-Z]', word)] # Display filtered words to observe words after tokenization filtered<jupyter_output><empty_output><jupyter_text>## 4. Stemming Stemming is the process by which we bring down a word from its different forms to the root word. This helps us establish meaning to different forms of the same words without having to deal with each form separately. For example, the words 'fishing', 'fished', and 'fisher' all get stemmed to the word 'fish'. Consider the following sentences: "Young William Wallace witnesses the treachery of Longshanks" ~ Gladiator "escapes to the city walls only to witness Cicero's death" ~ Braveheart Instead of building separate dictionary entries for both witnesses and witness, which mean the same thing outside of quantity, stemming them reduces them to 'wit'. There are different algorithms available for stemming such as the Porter Stemmer, Snowball Stemmer, etc. We shall use the Snowball Stemmer.<jupyter_code># Import the SnowballStemmer to perform stemming from nltk.stem.snowball import SnowballStemmer # Create an English language SnowballStemmer object stemmer = SnowballStemmer("english") # Print filtered to observe words without stemming print("Without stemming: ", filtered) # Stem the words from filtered and store in stemmed_words stemmed_words = [stemmer.stem(t) for t in filtered] # Print the stemmed_words to observe words after stemming print("After stemming: ", stemmed_words)<jupyter_output>Without stemming: ['Today', 'May', 'is', 'his', 'only', 'daughter', "'s", 'wedding'] After stemming: ['today', 'may', 'is', 'his', 'onli', 'daughter', "'s", 'wed'] <jupyter_text>## 5. Club together Tokenize & Stem We are now able to tokenize and stem sentences. But we may have to use the two functions repeatedly one after the other to handle a large amount of data, hence we can think of wrapping them in a function and passing the text to be tokenized and stemmed as the function argument. Then we can pass the new wrapping function, which shall perform both tokenizing and stemming instead of just tokenizing, as the tokenizer argument while creating the TF-IDF vector of the text. What difference does it make though? Consider the sentence from the plot of The Godfather: "Today (May 19, 2016) is his only daughter's wedding." If we do a 'tokenize-only' for this sentence, we have the following result: 'today', 'may', 'is', 'his', 'only', 'daughter', "'s", 'wedding' But when we do a 'tokenize-and-stem' operation we get: 'today', 'may', 'is', 'his', 'onli', 'daughter', "'s", 'wed' All the words are in their root form, which will lead to a better establishment of meaning as some of the non-root forms may not be present in the NLTK training corpus.<jupyter_code># Define a function to perform both stemming and tokenization def tokenize_and_stem(text): # Tokenize by sentence, then by word tokens = [word for sent in nltk.sent_tokenize(text) for word in nltk.word_tokenize(sent)] # Filter out raw tokens to remove noise filtered_tokens = [token for token in tokens if re.search('[a-zA-Z]', token)] # Stem the filtered_tokens stems = [stemmer.stem(t) for t in filtered_tokens] return stems words_stemmed = tokenize_and_stem("Today (May 19, 2016) is his only daughter's wedding.") print(words_stemmed)<jupyter_output>['today', 'may', 'is', 'his', 'onli', 'daughter', "'s", 'wed'] <jupyter_text>## 6. Create TfidfVectorizer Computers do not understand text. These are machines only capable of understanding numbers and performing numerical computation. Hence, we must convert our textual plot summaries to numbers for the computer to be able to extract meaning from them. One simple method of doing this would be to count all the occurrences of each word in the entire vocabulary and return the counts in a vector. Enter CountVectorizer. Consider the word 'the'. It appears quite frequently in almost all movie plots and will have a high count in each case. But obviously, it isn't the theme of all the movies! Term Frequency-Inverse Document Frequency (TF-IDF) is one method which overcomes the shortcomings of CountVectorizer. The Term Frequency of a word is the measure of how often it appears in a document, while the Inverse Document Frequency is the parameter which reduces the importance of a word if it frequently appears in several documents. For example, when we apply the TF-IDF on the first 3 sentences from the plot of The Wizard of Oz, we are told that the most important word there is 'Toto', the pet dog of the lead character. This is because the movie begins with 'Toto' biting someone due to which the journey of Oz begins! In simplest terms, TF-IDF recognizes words which are unique and important to any given document. Let's create one for our purposes.<jupyter_code># Import TfidfVectorizer to create TF-IDF vectors from sklearn.feature_extraction.text import TfidfVectorizer # Instantiate TfidfVectorizer object with stopwords and tokenizer # parameters for efficient processing of text tfidf_vectorizer = TfidfVectorizer(max_df=0.8, max_features=200000, min_df=0.2, stop_words='english', use_idf=True, tokenizer=tokenize_and_stem, ngram_range=(1,3))<jupyter_output><empty_output><jupyter_text>## 7. Fit transform TfidfVectorizer Once we create a TF-IDF Vectorizer, we must fit the text to it and then transform the text to produce the corresponding numeric form of the data which the computer will be able to understand and derive meaning from. To do this, we use the fit_transform() method of the TfidfVectorizer object. If we observe the TfidfVectorizer object we created, we come across a parameter stopwords. 'stopwords' are those words in a given text which do not contribute considerably towards the meaning of the sentence and are generally grammatical filler words. For example, in the sentence 'Dorothy Gale lives with her dog Toto on the farm of her Aunt Em and Uncle Henry', we could drop the words 'her' and 'the', and still have a similar overall meaning to the sentence. Thus, 'her' and 'the' are stopwords and can be conveniently dropped from the sentence. On setting the stopwords to 'english', we direct the vectorizer to drop all stopwords from a pre-defined list of English language stopwords present in the nltk module. Another parameter, ngram_range, defines the length of the ngrams to be formed while vectorizing the text.<jupyter_code># Fit and transform the tfidf_vectorizer with the "plot" of each movie # to create a vector representation of the plot summaries tfidf_matrix = tfidf_vectorizer.fit_transform([x for x in movies_df["plot"]]) print(tfidf_matrix.shape)<jupyter_output>(100, 564) <jupyter_text>## 8. Import KMeans and create clusters To determine how closely one movie is related to the other by the help of unsupervised learning, we can use clustering techniques. Clustering is the method of grouping together a number of items such that they exhibit similar properties. According to the measure of similarity desired, a given sample of items can have one or more clusters. A good basis of clustering in our dataset could be the genre of the movies. Say we could have a cluster '0' which holds movies of the 'Drama' genre. We would expect movies like Chinatown or Psycho to belong to this cluster. Similarly, the cluster '1' in this project holds movies which belong to the 'Adventure' genre (Lawrence of Arabia and the Raiders of the Lost Ark, for example). K-means is an algorithm which helps us to implement clustering in Python. The name derives from its method of implementation: the given sample is divided into K clusters where each cluster is denoted by the mean of all the items lying in that cluster. We get the following distribution for the clusters: <jupyter_code># Import k-means to perform clustering from sklearn.cluster import KMeans # Create a KMeans object with 5 clusters and save as km km = KMeans(n_clusters=5) # Fit the k-means object with tfidf_matrix km.fit(tfidf_matrix) clusters = km.labels_.tolist() # Create a column cluster to denote the generated cluster for each movie movies_df["cluster"] = clusters # Display number of films per cluster (clusters from 0 to 4) movies_df['cluster'].value_counts() <jupyter_output><empty_output><jupyter_text>## 9. Calculate similarity distance Consider the following two sentences from the movie The Wizard of Oz: "they find in the Emerald City" "they finally reach the Emerald City" If we put the above sentences in a CountVectorizer, the vocabulary produced would be "they, find, in, the, Emerald, City, finally, reach" and the vectors for each sentence would be as follows: 1, 1, 1, 1, 1, 1, 0, 0 1, 0, 0, 1, 1, 1, 1, 1 When we calculate the cosine angle formed between the vectors represented by the above, we get a score of 0.667. This means the above sentences are very closely related. Similarity distance is 1 - cosine similarity angle. This follows from that if the vectors are similar, the cosine of their angle would be 1 and hence, the distance between then would be 1 - 1 = 0. Let's calculate the similarity distance for all of our movies.<jupyter_code># Import cosine_similarity to calculate similarity of movie plots from sklearn.metrics.pairwise import cosine_similarity # Calculate the similarity distance similarity_distance = 1 - cosine_similarity(tfidf_matrix)<jupyter_output><empty_output><jupyter_text>## 10. Import Matplotlib, Linkage, and Dendrograms We shall now create a tree-like diagram (called a dendrogram) of the movie titles to help us understand the level of similarity between them visually. Dendrograms help visualize the results of hierarchical clustering, which is an alternative to k-means clustering. Two pairs of movies at the same level of hierarchical clustering are expected to have similar strength of similarity between the corresponding pairs of movies. For example, the movie Fargo would be as similar to North By Northwest as the movie Platoon is to Saving Private Ryan, given both the pairs exhibit the same level of the hierarchy. Let's import the modules we'll need to create our dendrogram.<jupyter_code># Import matplotlib.pyplot for plotting graphs import matplotlib.pyplot as plt # Configure matplotlib to display the output inline %matplotlib inline # Import modules necessary to plot dendrogram from scipy.cluster.hierarchy import linkage, dendrogram<jupyter_output><empty_output><jupyter_text>## 11. Create merging and plot dendrogram We shall plot a dendrogram of the movies whose similarity measure will be given by the similarity distance we previously calculated. The lower the similarity distance between any two movies, the lower their linkage will make an intercept on the y-axis. For instance, the lowest dendrogram linkage we shall discover will be between the movies, It's a Wonderful Life and A Place in the Sun. This indicates that the movies are very similar to each other in their plots.<jupyter_code># Create mergings matrix mergings = linkage(similarity_distance, method='complete') # Plot the dendrogram, using title as label column dendrogram_ = dendrogram(mergings, labels=[x for x in movies_df["title"]], leaf_rotation=90, leaf_font_size=16, ) # Adjust the plot fig = plt.gcf() _ = [lbl.set_color('r') for lbl in plt.gca().get_xmajorticklabels()] fig.set_size_inches(108, 21) # Show the plotted dendrogram plt.show()<jupyter_output><empty_output><jupyter_text>## 12. Which movies are most similar? We can now determine the similarity between movies based on their plots! To wrap up, let's answer one final question: which movie is most similar to the movie Braveheart?<jupyter_code># Answer the question ans = "Gladiator" print(ans)<jupyter_output>Gladiator
no_license
/Projects/Find Movie Similarity from Plot Summaries/Movie_Prediction.ipynb
Gttz/DataCamp_Courses
12
<jupyter_start><jupyter_text>#1. Preprocess your data so that you can feed it into ANN models. #2. Split your data into training and test sets.<jupyter_code>input_dim = 784 # 28*28 output_dim = nb_classes = 10 batch_size = 128 nb_epoch = 20 X_train = X_train.reshape(60000, input_dim) X_test = X_test.reshape(10000, input_dim) X_train = X_train.astype('float32') X_test = X_test.astype('float32') X_train /= 255 X_test /= 255 from tensorflow.keras.utils import to_categorical Y_train = to_categorical(y_train, nb_classes) Y_test = to_categorical(y_test, nb_classes) import matplotlib.pyplot as plt plt.figure(figsize=(20,10)) for i in range(1,21): plt.subplot(4,5,i) plt.imshow(X_train[120+i].reshape(28,28), cmap="gray") plt.title("Label of the image: {}".format(y_train[120 + i])) plt.tight_layout() plt.show()<jupyter_output><empty_output><jupyter_text>#3. Try different ANN models and train them on your training set.##3.1. Number of layers.<jupyter_code>from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense<jupyter_output><empty_output><jupyter_text>###1 Hidden Layer<jupyter_code># 1 Hidden Layer model_1 = Sequential() # our first dense layer model_1.add(Dense(128, input_shape=(784,), activation="relu")) # last layer is the output layer. model_1.add(Dense(10, activation="softmax")) # compile model model_1.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_1.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_1 = model_1.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_1[0]) print('Test accuracy:', score_1[1])<jupyter_output>Test score: 0.45044970512390137 Test accuracy: 0.8402000069618225 <jupyter_text>###2 Hidden Layers<jupyter_code># 2 Hidden Layers model_2 = Sequential() # our first dense layer model_2.add(Dense(128, input_shape=(784,), activation="relu")) # our second dense layer model_2.add(Dense(128, activation="relu")) # last layer is the output layer. model_2.add(Dense(10, activation="softmax")) # compile model model_2.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_2.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_2 = model_2.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_2[0]) print('Test accuracy:', score_2[1])<jupyter_output>Test score: 0.4185384213924408 Test accuracy: 0.8507999777793884 <jupyter_text>###3 Hidden Layers<jupyter_code># 3 Hidden Layers model_3 = Sequential() # our first dense layer model_3.add(Dense(128, input_shape=(784,), activation="relu")) # our second dense layer model_3.add(Dense(128, activation="relu")) # our third dense layer model_3.add(Dense(128, activation="relu")) # last layer is the output layer. model_3.add(Dense(10, activation="softmax")) # compile model model_3.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_3.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_3 = model_3.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_3[0]) print('Test accuracy:', score_3[1])<jupyter_output>Test score: 0.4012981653213501 Test accuracy: 0.8551999926567078 <jupyter_text>###4 Hidden Layers<jupyter_code># 4 Hidden Layers model_4 = Sequential() # our first dense layer model_4.add(Dense(128, input_shape=(784,), activation="relu")) # our second dense layer model_4.add(Dense(128, activation="relu")) # our third dense layer model_4.add(Dense(128, activation="relu")) # our fourth dense layer model_4.add(Dense(128, activation="relu")) # last layer is the output layer. model_4.add(Dense(10, activation="softmax")) # compile model model_4.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_4.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_4 = model_4.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_4[0]) print('Test accuracy:', score_4[1])<jupyter_output>Test score: 0.39575937390327454 Test accuracy: 0.8571000099182129 <jupyter_text>###5. Hidden Layers<jupyter_code># 5 Hidden Layers model_5 = Sequential() # our first dense layer model_5.add(Dense(128, input_shape=(784,), activation="relu")) # our second dense layer model_5.add(Dense(128, activation="relu")) # our third dense layer model_5.add(Dense(128, activation="relu")) # our fourth dense layer model_5.add(Dense(128, activation="relu")) # our fifth dense layer model_5.add(Dense(128, activation="relu")) # last layer is the output layer. model_5.add(Dense(10, activation="softmax")) # compile model model_5.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_5.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_5 = model_5.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_5[0]) print('Test accuracy:', score_5[1]) layers = [1, 2, 3, 4, 5] scores = [score_1[1], score_2[1], score_3[1], score_4[1], score_5[1]] plt.figure(figsize=(10,6)) plt.plot(layers,scores) plt.title("Model Score vs Hidden Layers") plt.xlabel("Hidden Layers") plt.xticks(layers) plt.ylabel("Test Accuracy") plt.grid(alpha=0.5) plt.show()<jupyter_output><empty_output><jupyter_text>###We move forward using 4 hidden layers.##3.2 Activation Functions<jupyter_code># ReLU activation model_relu = Sequential() # our first dense layer model_relu.add(Dense(128, input_shape=(784,), activation="relu")) # our second dense layer model_relu.add(Dense(128, activation="relu")) # our third dense layer model_relu.add(Dense(128, activation="relu")) # our fourth dense layer model_relu.add(Dense(128, activation="relu")) # last layer is the output layer. model_relu.add(Dense(10, activation="softmax")) # compile model model_relu.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_relu.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_relu = model_relu.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_relu[0]) print('Test accuracy:', score_relu[1]) model_sigm = Sequential() # our first dense layer model_sigm.add(Dense(128, input_shape=(784,), activation="sigmoid")) # our second dense layer model_sigm.add(Dense(128, activation="sigmoid")) # our third dense layer model_sigm.add(Dense(128, activation="sigmoid")) # our fourth dense layer model_sigm.add(Dense(128, activation="sigmoid")) # last layer is the output layer. model_sigm.add(Dense(10, activation="softmax")) # compile model model_sigm.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_sigm.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_sigm = model_sigm.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_sigm[0]) print('Test accuracy:', score_sigm[1]) model_tanh = Sequential() # our first dense layer model_tanh.add(Dense(128, input_shape=(784,), activation="tanh")) # our second dense layer model_tanh.add(Dense(128, activation="tanh")) # our third dense layer model_tanh.add(Dense(128, activation="tanh")) # our fourth dense layer model_tanh.add(Dense(128, activation="tanh")) # last layer is the output layer. model_tanh.add(Dense(10, activation="softmax")) # compile model model_tanh.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_tanh.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_tanh = model_tanh.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_tanh[0]) print('Test accuracy:', score_tanh[1]) model_swish = Sequential() # our first dense layer model_swish.add(Dense(128, input_shape=(784,), activation="swish")) # our second dense layer model_swish.add(Dense(128, activation="swish")) # our third dense layer model_swish.add(Dense(128, activation="swish")) # our fourth dense layer model_swish.add(Dense(128, activation="swish")) # last layer is the output layer. model_swish.add(Dense(10, activation="softmax")) # compile model model_swish.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_swish.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_swish = model_swish.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_swish[0]) print('Test accuracy:', score_swish[1]) activation = ['ReLU', 'Sigmoid', 'tanh', 'Swish'] scores = [score_relu[1], score_sigm[1], score_tanh[1], score_swish[1]] plt.figure(figsize=(10,6)) plt.bar(activation,scores) plt.title("Model Score vs Activation Function") plt.xlabel("Activation Function") plt.ylabel("Test Accuracy") plt.grid(alpha=0.5, axis= 'y') plt.show()<jupyter_output><empty_output><jupyter_text>###ReLU and Hyperbolic Tanger perform very similarly. We elect to use ReLU since it's non-saturating.##3.3 Number of Neurons in the Layers<jupyter_code># 256 in the first layer with a 1/2 reduction after model_256r = Sequential() # our first dense layer model_256r.add(Dense(256, input_shape=(784,), activation="relu")) # our second dense layer model_256r.add(Dense(128, activation="relu")) # our third dense layer model_256r.add(Dense(64, activation="relu")) # our fourth dense layer model_256r.add(Dense(32, activation="relu")) # last layer is the output layer. model_256r.add(Dense(10, activation="softmax")) # compile model model_256r.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_256r.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_256r = model_256r.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_256r[0]) print('Test accuracy:', score_256r[1]) # 256 in each layer model_256c = Sequential() # our first dense layer model_256c.add(Dense(256, input_shape=(784,), activation="relu")) # our second dense layer model_256c.add(Dense(256, activation="relu")) # our third dense layer model_256c.add(Dense(256, activation="relu")) # our fourth dense layer model_256c.add(Dense(256, activation="relu")) # last layer is the output layer. model_256c.add(Dense(10, activation="softmax")) # compile model model_256c.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_256c.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_256c = model_256c.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_256c[0]) print('Test accuracy:', score_256c[1]) # 512 in the first layer with a 1/2 reduction after model_512r = Sequential() # our first dense layer model_512r.add(Dense(512, input_shape=(784,), activation="relu")) # our second dense layer model_512r.add(Dense(256, activation="relu")) # our third dense layer model_512r.add(Dense(128, activation="relu")) # our fourth dense layer model_512r.add(Dense(64, activation="relu")) # last layer is the output layer. model_512r.add(Dense(10, activation="softmax")) # compile model model_512r.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_512r.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_512r = model_512r.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_512r[0]) print('Test accuracy:', score_512r[1]) # 512 in each layer model_512c = Sequential() # our first dense layer model_512c.add(Dense(512, input_shape=(784,), activation="relu")) # our second dense layer model_512c.add(Dense(512, activation="relu")) # our third dense layer model_512c.add(Dense(512, activation="relu")) # our fourth dense layer model_512c.add(Dense(512, activation="relu")) # last layer is the output layer. model_512c.add(Dense(10, activation="softmax")) # compile model model_512c.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_512c.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_512c = model_swish.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_512c[0]) print('Test accuracy:', score_512c[1]) # 256, 128, 128, 64 model_256t = Sequential() # our first dense layer model_256t.add(Dense(256, input_shape=(784,), activation="relu")) # our second dense layer model_256t.add(Dense(128, activation="relu")) # our third dense layer model_256t.add(Dense(128, activation="relu")) # our fourth dense layer model_256t.add(Dense(64, activation="relu")) # last layer is the output layer. model_256t.add(Dense(10, activation="softmax")) # compile model model_256t.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_256t.fit(X_train, Y_train, batch_size=batch_size, epochs=20, verbose=1) score_256t = model_256t.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_256t[0]) print('Test accuracy:', score_256t[1])<jupyter_output>Test score: 0.37574777007102966 Test accuracy: 0.8639000058174133 <jupyter_text>###The best combination of training speed, and test accuracy came with hidden layers of 256, 128, 64 and 32 neurons.##3.4 Batch Size<jupyter_code># batch size of 1024 model_1024 = Sequential() # our first dense layer model_1024.add(Dense(256, input_shape=(784,), activation="relu")) # our second dense layer model_1024.add(Dense(128, activation="relu")) # our third dense layer model_1024.add(Dense(64, activation="relu")) # our fourth dense layer model_1024.add(Dense(32, activation="relu")) # last layer is the output layer. model_1024.add(Dense(10, activation="softmax")) # compile model model_1024.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_1024.fit(X_train, Y_train, batch_size=1024, epochs=20, verbose=1) score_1024 = model_1024.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_1024[0]) print('Test accuracy:', score_1024[1]) # batch size of 256 model_256 = Sequential() # our first dense layer model_256.add(Dense(256, input_shape=(784,), activation="relu")) # our second dense layer model_256.add(Dense(128, activation="relu")) # our third dense layer model_256.add(Dense(64, activation="relu")) # our fourth dense layer model_256.add(Dense(32, activation="relu")) # last layer is the output layer. model_256.add(Dense(10, activation="softmax")) # compile model model_256.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_256.fit(X_train, Y_train, batch_size=256, epochs=20, verbose=1) score_256 = model_256.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_256[0]) print('Test accuracy:', score_256[1]) # batch size of 64 model_64 = Sequential() # our first dense layer model_64.add(Dense(256, input_shape=(784,), activation="relu")) # our second dense layer model_64.add(Dense(128, activation="relu")) # our third dense layer model_64.add(Dense(64, activation="relu")) # our fourth dense layer model_64.add(Dense(32, activation="relu")) # last layer is the output layer. model_64.add(Dense(10, activation="softmax")) # compile model model_64.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_64.fit(X_train, Y_train, batch_size=64, epochs=20, verbose=1) score_64 = model_64.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_64[0]) print('Test accuracy:', score_64[1]) # batch size of 16 model_16 = Sequential() # our first dense layer model_16.add(Dense(256, input_shape=(784,), activation="relu")) # our second dense layer model_16.add(Dense(128, activation="relu")) # our third dense layer model_16.add(Dense(64, activation="relu")) # our fourth dense layer model_16.add(Dense(32, activation="relu")) # last layer is the output layer. model_16.add(Dense(10, activation="softmax")) # compile model model_16.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy']) # setting verbose=1 prints out some results after each epoch model_16.fit(X_train, Y_train, batch_size=16, epochs=20, verbose=1) score_16 = model_16.evaluate(X_test, Y_test, verbose=0) print('Test score:', score_16[0]) print('Test accuracy:', score_16[1])<jupyter_output>Test score: 0.335713654756546 Test accuracy: 0.8852999806404114 <jupyter_text>###Batch size of 16 takes a very long time to train, and the difference between training and test accuracy is almost 4%, indicating some significant over-fitting. We settle on batch size of 64.#General Discussion This dataset proved to be much more difficult to achieve a high accuracy on than the handwritten digits set. While tuning the number of layers, neurons in each layer, and batch size we ran into the same type of trade off. Increasing the complexity of the model, and its evaluation lead to much longer training times, as there were more parameters/equations to solve. The increased complexity often yielded higher training accuracies, but there was a point where the corresponding increase in test accuracy was negligible, and we saw overfitting in action. In the end, we settled on a model with the following characteristics: 4 Hidden Layers: 256, 128, 64 and 32 neurons in each. ReLU Activation Function Batch size of 64 Training Accuracy: 90.06% Test Accuracy: 87.29%<jupyter_code><jupyter_output><empty_output>
no_license
/deep-learning-challenge/deep_learning_challenge.ipynb
ADEnnaco/thinkful-coursework
11
<jupyter_start><jupyter_text># 新做的<jupyter_code>data = [] label2 = [] label3 = [] for i in range(1,25): a = np.load('/content/drive/My Drive/Newdata/ordered_by_patient/150_6_116_116_' + str(i) +'_flipped.npy') c = np.load('/content/drive/My Drive/Newdata/ordered_by_patient/two_labels/' + str(i) + '.npy') #两个label d = np.load('/content/drive/My Drive/Newdata/ordered_by_patient/percent_output_' + str(i) + '.npy') #三个label data.append(a) label2.append(c) label3.append(d) Data, Label2, Label3 = shuffle(data, label2, label3, random_state = 20) P = np.array(Data) P = np.reshape(P,(3600,6,116,116,1)) L2 = np.array(Label2) L2 = np.reshape(L2,(3600,2)) L3 = np.array(Label3) L3 = np.reshape(L3,(3600,3)) print(P.shape) print(L2.shape) print(L3.shape) #输入数据 train_data = P[:2700] test_data = P[2700:] # 全部打乱 没有按照病人 Data, Label2, Label3 = shuffle(Data, Label2, Label3, random_state = 20) # 2 label train_label = L2[:2700] test_label = L2[2700:] print(train_data.shape) print(train_label.shape) # 3 label train_label = L3[:2700] test_label = L3[2700:] print(train_data.shape) print(train_label.shape) train_data, train_label = shuffle(train_data, train_label, random_state = 20) # count = 0 # for i in range(24): # for j in range(150): # # if Label3[i,j,0] + Label3[i,j,1] + Label3[i,j,2] == 1: # if L3[i,j,0] + L3[i,j,1] + L3[i,j,2] == 1: # count+=1 # print(count) count = 0 for i in range(3600): if L3[i,0] + L3[i,1] + L3[i,2] == 1: count+=1 print(count) # print(L3) # # data1 = np.load('/content/drive/My Drive/Newdata/original/allpre.npy') # # data2 = np.load('/content/drive/My Drive/Newdata/original/allpost.npy') # # data3 = np.load('/content/drive/My Drive/Newdata/original/allfu.npy') # # data = np.concatenate((data1,data2,data3), axis = 0) # data0 = np.load('/content/drive/My Drive/Newdata/original/allcontrol.npy') # data1 = np.load('/content/drive/My Drive/Newdata/changed/allpre.npy') # data2 = np.load('/content/drive/My Drive/Newdata/changed/allpost.npy') # data3 = np.load('/content/drive/My Drive/Newdata/changed/allfu.npy') # data = np.concatenate((data0,data1,data2,data3), axis = 0) # label = np.load('/content/drive/My Drive/Newdata/original/4800.npy') # data, label = shuffle(data,label,random_state=20) # # data, label = shuffle(data,label[1200:],random_state=20) # # np.save("/content/drive/My Drive/Newdata/all4800",Data) # np.save("/content/drive/My Drive/Newdata/shuffle_changedin4800",data) # np.save("/content/drive/My Drive/Newdata/shuffle_changedout4800",label) def Model(): model = Sequential() initializer = tf.keras.initializers.HeNormal() model.add(Conv3D(32, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer, input_shape=(6,116,116, 1))) model.add(MaxPooling3D(pool_size=(1, 2, 2))) model.add(Dropout(0.3)) # model.add(Conv3D(64, kernel_size=(2, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(Dropout(0.3)) model.add(Conv3D(64, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) model.add(MaxPooling3D(pool_size=(1, 2, 2))) model.add(Dropout(0.3)) model.add(Conv3D(128, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) model.add(Dropout(0.3)) # model.add(Conv3D(128, kernel_size=(2, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(Dropout(0.3)) model.add(Conv3D(128, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) model.add(MaxPooling3D(pool_size=(1, 2, 2))) model.add(Dropout(0.3)) # model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(Dropout(0.3)) # model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(Dropout(0.3)) model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) model.add(MaxPooling3D(pool_size=(1, 2, 2))) model.add(Dropout(0.3)) # model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(1, 2, 2))) # model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(1, 2, 2))) # model.add(Dropout(0.3)) # model.add(Conv3D(512, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(1, 2, 2))) # model.add(Dropout(0.3)) # model.add(Conv3D(1024, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(1, 2, 2))) # model.add(Dropout(0.3)) # model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(Conv3D(32, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(Conv3D(32, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(Conv3D(32, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(2, 2, 2))) # model.add(BatchNormalization()) # model.add(Dropout(0.3)) # model.add(Conv3D(64, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(2, 2, 2))) # model.add(BatchNormalization()) # model.add(Dropout(0.5)) # model.add(Conv3D(128, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(2, 2, 2))) # model.add(Dropout(0.5)) # model.add(BatchNormalization()) # model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer=initializer)) # model.add(MaxPooling3D(pool_size=(2, 2, 2))) model.add(BatchNormalization()) # model.add(Dropout(0.5)) model.add(Flatten()) model.add(Dense(1024, activation='relu', kernel_initializer=initializer)) model.add(Dense(3)) model.summary() from keras.optimizers import RMSprop,Adam,SGD model.compile(optimizer = SGD(learning_rate=0.0001, momentum=0.9), loss = 'mean_squared_error', metrics = ['mae']) return model model = Model() # model = Model() num_epochs = 100 batchsize = 32 history = model.fit(train_data, train_label, epochs=num_epochs, batch_size=batchsize, verbose=1) mse = history.history['loss'] mae = history.history['mae'] # mse1 = history1.history['loss'] # mae1 = history1.history['mae'] # mse2 = history2.history['loss'] # mae2 = history2.history['mae'] # mse3 = history3.history['loss'] # mae3 = history3.history['mae'] # mse4 = history4.history['loss'] # mae4 = history4.history['mae'] # mse5 = history5.history['loss'] # mae5 = history5.history['mae'] # mse6 = history6.history['loss'] # mae6 = history6.history['mean_absolute_error'] # mse7 = history7.history['loss'] # mae7 = history7.history['mean_absolute_error'] # mse8 = history8.history['loss'] # mae8 = history8.history['mean_absolute_error'] # model = ZF_Net() # model1 = VGG16() # model2 = ResNet50() # model3 = Xception() # model4 = DenseNet121() # model5 = MobileNet() # model6 = get_model9() epochs = range(1,len(mse) +1) plt.plot(epochs, mse, 'b', label='Proposed Net') plt.plot(epochs, mae, 'b--', label='mae') # plt.plot(epochs, mse1, 'r', label='VGG16') # plt.plot(epochs, mae1, 'r--', label='mae1') # plt.plot(epochs, mse2, 'g', label='ResNet50') # plt.plot(epochs, mae2, 'g--', label='mae2') # plt.plot(epochs, mse3, 'c', label='Xception') # plt.plot(epochs, mae3, 'c--', label='mae3') # plt.plot(epochs, mse4, 'm', label='DenseNet121') # plt.plot(epochs, mae4, 'm--', label='mae4') # plt.plot(epochs, mse5, 'y', label='MobileNet') # plt.plot(epochs, mae5, 'y--', label='mae5') # plt.plot(epochs, mse6, 'k', label='model9') # plt.plot(epochs, mae6, 'k--', label='mae6') # plt.plot(epochs, mse7, '#9467bd', label='mse7') # plt.plot(epochs, mae7, '#9467bd',linestyle='dashed', label='mae7') # plt.plot(epochs, mse8, '#e377c2', label='mse8') # plt.plot(epochs, mae8, '#e377c2',linestyle='dashed', label='mae8') plt.legend(bbox_to_anchor=(0.8, 1), loc='upper left', borderaxespad=0.) # plt.title('loss') plt.xlabel('Epochs') plt.ylabel('Loss & MAE') # plt.show() # plt.savefig('/content/drive/My Drive/images/pretrained_models_3_wo.png' ,dpi=1200) newx = test_data m = model.predict(newx) * 100 # m1 = model1.predict(newx) * 100 # m2 = model2.predict(newx) * 100 # m3 = model3.predict(newx) * 100 # m4 = model4.predict(newx) * 100 # m5 = model5.predict(newx) * 100 print(m.shape) print(test_label[0]) print(m[0]) c1 = test_label[:,0] * 100# 对测试集的真实数值c1进行大小排列 c2 = test_label[:,1] * 100# 对测试集的真实数值c2进行大小排列 c3 = test_label[:,2] * 100# 对测试集的真实数值c3进行大小排列 # print(y_data_c1) index1 = np.argsort(c1) #c1提取出按照从小到大的对应index index2 = np.argsort(c2) #c2提取出按照从小到大的对应index index3 = np.argsort(c3) #c3提取出按照从小到大的对应index c1.sort() c2.sort() c3.sort() # print(index1) yc1 = [] yc2 = [] yc3 = [] # y1c1 = [] # y1c2 = [] # y1c3 = [] # y2c1 = [] # y2c2 = [] # y2c3 = [] # y3c1 = [] # y3c2 = [] # y3c3 = [] # y4c1 = [] # y4c2 = [] # y4c3 = [] # y5c1 = [] # y5c2 = [] # y5c3 = [] for i in range(900): yc1.append([m[i,0],index1[i]]) yc2.append([m[i,1],index2[i]]) yc3.append([m[i,2],index3[i]]) # y1c1.append([m1[i,0],index1[i]]) # y1c2.append([m1[i,1],index2[i]]) # y1c3.append([m1[i,2],index3[i]]) # y2c1.append([m2[i,0],index1[i]]) # y2c2.append([m2[i,1],index2[i]]) # y2c3.append([m2[i,2],index3[i]]) # y3c1.append([m3[i,0],index1[i]]) # y3c2.append([m3[i,1],index2[i]]) # y3c3.append([m3[i,2],index3[i]]) # y4c1.append([m4[i,0],index1[i]]) # y4c2.append([m4[i,1],index2[i]]) # y4c3.append([m4[i,2],index3[i]]) # y5c1.append([m5[i,0],index1[i]]) # y5c2.append([m5[i,1],index2[i]]) # y5c3.append([m5[i,2],index3[i]]) yc1.sort(key = lambda x: x[1]) yc2.sort(key = lambda x: x[1]) yc3.sort(key = lambda x: x[1]) # y1c1.sort(key = lambda x: x[1]) # y1c2.sort(key = lambda x: x[1]) # y1c3.sort(key = lambda x: x[1]) # y2c1.sort(key = lambda x: x[1]) # y2c2.sort(key = lambda x: x[1]) # y2c3.sort(key = lambda x: x[1]) # y3c1.sort(key = lambda x: x[1]) # y3c2.sort(key = lambda x: x[1]) # y3c3.sort(key = lambda x: x[1]) # y4c1.sort(key = lambda x: x[1]) # y4c2.sort(key = lambda x: x[1]) # y4c3.sort(key = lambda x: x[1]) # y5c1.sort(key = lambda x: x[1]) # y5c2.sort(key = lambda x: x[1]) # y5c3.sort(key = lambda x: x[1]) samples = 900 x_data = range(samples) # y_data_c1 = test_label[:,0] # y_data_c2 = test_label[:,1] # y_data_c3 = test_label[:,2] yc1 = np.array(yc1) yc2 = np.array(yc2) yc3 = np.array(yc3) # y1c1 = np.array(y1c1) # y1c2 = np.array(y1c2) # y1c3 = np.array(y1c3) # y2c1 = np.array(y2c1) # y2c2 = np.array(y2c2) # y2c3 = np.array(y2c3) # y3c1 = np.array(y3c1) # y3c2 = np.array(y3c2) # y3c3 = np.array(y3c3) # y4c1 = np.array(y4c1) # y4c2 = np.array(y4c2) # y4c3 = np.array(y4c3) # y5c1 = np.array(y5c1) # y5c2 = np.array(y5c2) # y5c3 = np.array(y5c3) # y_predict_c1 = m[:,0] # y_predict_c2 = m[:,1] # y_predict_c3 = m[:,2] # y_predict1_c1 = m1[:,0] # y_predict1_c2 = m1[:,1] # y_predict1_c3 = m1[:,2] # y_predict2_c1 = m2[:,0] # y_predict2_c2 = m2[:,1] # y_predict2_c3 = m2[:,2] # y_predict3_c1 = m3[:,0] # y_predict3_c2 = m3[:,1] # y_predict3_c3 = m3[:,2] # y_predict4_c1 = m4[:,0] # y_predict4_c2 = m4[:,1] # y_predict4_c3 = m4[:,2] # y_predict5_c1 = m5[:,0] # y_predict5_c2 = m5[:,1] # y_predict5_c3 = m5[:,2] fig, (predict_c1, predict_c2, predict_c3) = plt.subplots(3,figsize=(20,5)) fig.suptitle('3 visual conditions') predict_c1.plot(x_data, c1[:samples] , 'c', label='truth_c1') predict_c2.plot(x_data, c2[:samples] , 'c', label='truth_c2') predict_c3.plot(x_data, c3[:samples] , 'c', label='truth_c3') predict_c1.plot(x_data, yc1[:samples,0] , 'r', label='My proposed_c1') predict_c2.plot(x_data, yc2[:samples,0], 'r', label='My proposed_c2') predict_c3.plot(x_data, yc3[:samples,0], 'r', label='My proposed_c3') # predict_c1.plot(x_data, y1c1[:samples,0] , 'b', label='VGG16_c1') # predict_c2.plot(x_data, y1c2[:samples,0] , 'b', label='VGG16_c2') # predict_c3.plot(x_data, y1c3[:samples,0] , 'b', label='VGG16_c3') # predict_c1.plot(x_data, y2c1[:samples,0] , 'k', label='ResNet50_c1') # predict_c2.plot(x_data, y2c2[:samples,0] , 'k', label='ResNet50_c2') # predict_c3.plot(x_data, y2c3[:samples,0] , 'k', label='ResNet50_c3') # predict_c1.plot(x_data, y3c1[:samples,0] , 'm', label='Xception_c1') # predict_c2.plot(x_data, y3c2[:samples,0] , 'm', label='Xception_c2') # predict_c3.plot(x_data, y3c3[:samples,0] , 'm', label='Xception_c3') # predict_c1.plot(x_data, y4c1[:samples,0] , 'y', label='DenseNet121_c1') # predict_c2.plot(x_data, y4c2[:samples,0] , 'y', label='DenseNet121_c2') # predict_c3.plot(x_data, y4c3[:samples,0] , 'y', label='DenseNet121_c3') # predict_c1.plot(x_data, y5c1[:samples,0] , 'g', label='MobileNet_c1') # predict_c2.plot(x_data, y5c2[:samples,0] , 'g', label='MobileNet_c2') # predict_c3.plot(x_data, y5c3[:samples,0] , 'g', label='MobileNet_c3') predict_c1.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) predict_c2.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) predict_c3.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) # plt.plot(x_data, y_data_c1 , 'c', label='c1') # plt.plot(x_data, y_data_c2 , 'r', label='c2') # plt.plot(x_data, y_data_c3 , 'b', label='c3') # plt.plot(x_data, y_predict1_c1 , 'c--', label='predict1_c1') # plt.plot(x_data, y_predict1_c2 , 'r--', label='predict1_c2') # plt.plot(x_data, y_predict1_c3 , 'b--', label='predict1_c3') # plt.plot(x_data, y_predict2_c1 , 'c:', label='predict2_c1') # plt.plot(x_data, y_predict2_c2 , 'r:', label='predict2_c2') # plt.plot(x_data, y_predict2_c3 , 'b:', label='predict2_c3') # plt.plot(x_data, y_predict3_c1 , 'c-.', label='predict3_c1') # plt.plot(x_data, y_predict3_c2 , 'r-.', label='predict3_c2') # plt.plot(x_data, y_predict3_c3 , 'b-.', label='predict3_c3') # plt.legend(loc="upper right",fontsize = 'small',bbox_to_anchor=(0.5, -0.05)) # plt.title('loss') # plt.xlabel('Epochs') # plt.ylabel('Loss') # plt.show() # fig.savefig('/content/drive/My Drive/33333333_wo.png' ,dpi=800) def calculate_mse(predict): #对比三种视觉情况 每个model的mse y_data_c1 = test_label[:,0] y_data_c2 = test_label[:,1] y_data_c3 = test_label[:,2] y_predict1_c1 = predict[:,0]/100 y_predict1_c2 = predict[:,1]/100 y_predict1_c3 = predict[:,2]/100 tmp1 = 0 tmp2 = 0 tmp3 = 0 for i in range(900): tmp1 = tmp1 + (y_data_c1[i]- y_predict1_c1[i])**2 tmp2 = tmp2 + (y_data_c2[i]- y_predict1_c2[i])**2 tmp3 = tmp3 + (y_data_c3[i]- y_predict1_c3[i])**2 c1 = tmp1/900 c2 = tmp2/900 c3 = tmp3/900 # c1 = K.mean(K.square( y_data_c1- y_predict1_c1), axis=-1) # c2 = K.mean(K.square( y_data_c2- y_predict1_c2), axis=-1) # c3 = K.mean(K.square( y_data_c3- y_predict1_c3), axis=-1) return c1, c2, c3 r = r2_score(test_label, m/100, multioutput='raw_values') # r1 = r2_score(test_label, m1/100, multioutput='raw_values') # r2 = r2_score(test_label, m2/100, multioutput='raw_values') # r3 = r2_score(test_label, m3/100, multioutput='raw_values') # r4 = r2_score(test_label, m4/100, multioutput='raw_values') # r5 = r2_score(test_label, m5/100, multioutput='raw_values') print(r) # print(r1) # print(r2) # print(r3) # print(r4) # print(r5) Model = calculate_mse(m) # Model1 = calculate_mse(m1) # Model2 = calculate_mse(m2) # Model3 = calculate_mse(m3) # Model4 = calculate_mse(m4) # Model5 = calculate_mse(m5) print(Model) # print(Model1) # print(Model2) # print(Model3) # print(Model4) # print(Model5) data = np.load('/content/drive/My Drive/Newdata/shuffle_changedin3600.npy') label = np.load('/content/drive/My Drive/Newdata/shuffle_changedout3600.npy') data = np.reshape(data,(3600,30,116,116,1)) split_rate= 0.7 split = int(split_rate*3600) train_data = data[:split] train_label = label[:split] test_data = data[split:] test_label = label[split:] def k_fold(k,train_data,train_targets): num_val_samples = len(train_data)//k val_mses = [] val_maes = [] mse_History = [] mae_History = [] num_epochs = 50 batchsize = 6 R2 = [] Comparision = [] for i in range(k): print('processing fold #', i) val_data = train_data[i * num_val_samples: (i + 1) * num_val_samples] val_targets = train_targets[i * num_val_samples: (i + 1) * num_val_samples] partial_train_data = np.concatenate([train_data[:i * num_val_samples],train_data[(i + 1) * num_val_samples:]], axis=0) partial_train_targets = np.concatenate([train_targets[:i * num_val_samples],train_targets[(i + 1) * num_val_samples:]], axis=0) model = Model() history = model.fit(partial_train_data, partial_train_targets, epochs=num_epochs, batch_size=batchsize, verbose=0) comparision1 = (model.predict(test_data))*100 Test_label = test_label*100 Comparision.append(comparision1) # r2_score(Test_label, comparision1, multioutput='raw_values') R2.append(r2_score(Test_label, comparision1, multioutput='raw_values')) # val_mse, val_mae = model.evaluate(val_data, val_targets, verbose=0) # mse_history = history.history['loss'] # mae_history = history.history['mean_absolute_error'] # val_mses.append(val_mse) # val_maes.append(val_mae) # mse_History.append(mse_history) # mae_History.append(mae_history) # return val_mses, val_maes, mse_History, mae_History return R2,Comparision R2_score, Predict_results = k_fold(5,train_data,train_label) # average_mae_history = [(np.mean([x[i] for x in mae_history]))*100 for i in range(50)] # average_mse_history = [(np.mean([x[i] for x in mse_history]))*100 for i in range(50)] # #model 3Dcnn # epochs = range(1,len(average_mse_history)+1) # plt.plot(epochs, average_mae_history,'c', label='mae' ) # plt.plot(epochs, average_mse_history,'m', label='mse' ) # plt.xlabel('Epochs') # plt.ylabel('Validation') # plt.show() data = np.load('/content/drive/My Drive/Newdata/shuffle_changedin4800.npy') label = np.load('/content/drive/My Drive/Newdata/shuffle_changedout4800.npy') data = np.reshape(data,(4800,30,116,116,1)) split_rate= 0.7 split = int(split_rate*4800) train_data = data[:split] train_label = label[:split] test_data = data[split:] test_label = label[split:] num_epochs = 50 batchsize = 6 model = Model() history = model.fit(train_data, train_label, epochs=num_epochs, batch_size=batchsize, verbose=0) # mse_history = history.history['loss'] # mae_history = history.history['mean_absolute_error'] comparision1 = (model.predict(test_data))*100 x_data = range(1080) Test_label = test_label*100 y_data_c1 = Test_label[:,0] y_data_c2 = Test_label[:,1] y_data_c3 = Test_label[:,2] y_predict1_c1 = comparision1[:,0] y_predict1_c2 = comparision1[:,1] y_predict1_c3 = comparision1[:,2] fig, (predict_c1, predict_c2, predict_c3) = plt.subplots(3,figsize=(18,6)) fig.suptitle('3 visual conditions_changed') # predict_c1.axes([0, 0.6, 1, 1]) # plt.subplots(figsize=(50, 10)) # plt.figure(figsize=(30,10) # fig, ax = plt.subplots(figsize=(20, 10)) # predict_c1.figure(figsize=(30,10) # bins = np.linspace(-1, 1, 21) #横坐标起始和结束值,分割成21份 # predict_c1.figure(figsize=(13,5)) #图像大小 # predict_c1.xticks(bins) #设置x轴 # predict_c1.xlim(-1, 1) #x轴开始和结束位置 # f = plt.figure(figsize=(20, 2)) # ax = f.add_subplot(111) # predict_c1.figure(figsize=(13,5)) # predict_c1.autoscale(enable=True, axis='x', tight=True)#plt.axis('tight') # predict_c1 predict_c1.plot(x_data, y_data_c1 , 'r', label='truth_c1') predict_c2.plot(x_data, y_data_c2 , 'r', label='truth_c2') predict_c3.plot(x_data, y_data_c3 , 'r', label='truth_c3') predict_c1.plot(x_data, y_predict1_c1 , 'g', label='predict1_c1') predict_c2.plot(x_data, y_predict1_c2 , 'g', label='predict1_c2') predict_c3.plot(x_data, y_predict1_c3 , 'g', label='predict1_c3') predict_c1.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) predict_c2.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) predict_c3.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) plt.savefig('/content/drive/My Drive/changed3600_50epo_0.7.png' ,dpi=1200) from sklearn.metrics import r2_score r2_score(Test_label, comparision1, multioutput='raw_values') <jupyter_output><empty_output><jupyter_text>#以下不要动系列 都是以前做的东西<jupyter_code>def Model(): model = Sequential() model.add(Conv3D(32, kernel_size=(2, 3, 3), activation='relu', kernel_initializer='he_uniform', input_shape=(30,116,116, 1))) model.add(MaxPooling3D(pool_size=(2, 2, 2))) model.add(BatchNormalization()) # model.add(Dropout(0.3)) model.add(Conv3D(64, kernel_size=(2, 3, 3), activation='relu', kernel_initializer='he_uniform')) model.add(MaxPooling3D(pool_size=(2, 2, 2))) model.add(BatchNormalization()) # model.add(Dropout(0.5)) model.add(Conv3D(128, kernel_size=(2, 3, 3), activation='relu', kernel_initializer='he_uniform')) model.add(MaxPooling3D(pool_size=(2, 2, 2))) # model.add(Dropout(0.5)) model.add(BatchNormalization()) model.add(Conv3D(256, kernel_size=(1, 3, 3), activation='relu', kernel_initializer='he_uniform')) model.add(MaxPooling3D(pool_size=(2, 2, 2))) model.add(BatchNormalization()) # model.add(Dropout(0.5)) model.add(Flatten()) model.add(Dense(256, activation='relu', kernel_initializer='he_uniform')) model.add(Dense(3, activation='softmax')) model.summary() from keras.optimizers import RMSprop,Adam model.compile(optimizer = Adam(lr=0.00001), loss = 'mean_squared_error', metrics = ['mae']) return model data = np.load('/content/drive/My Drive/Newdata/shuffle_changedin3600.npy') label = np.load('/content/drive/My Drive/Newdata/shuffle_changedout3600.npy') data = np.reshape(data,(3600,30,116,116,1)) split_rate= 0.7 split = int(split_rate*3600) train_data = data[:split] train_label = label[:split] test_data = data[split:] test_label = label[split:] def k_fold(k,train_data,train_targets): num_val_samples = len(train_data)//k val_mses = [] val_maes = [] mse_History = [] mae_History = [] num_epochs = 50 batchsize = 6 R2 = [] Comparision = [] for i in range(k): print('processing fold #', i) val_data = train_data[i * num_val_samples: (i + 1) * num_val_samples] val_targets = train_targets[i * num_val_samples: (i + 1) * num_val_samples] partial_train_data = np.concatenate([train_data[:i * num_val_samples],train_data[(i + 1) * num_val_samples:]], axis=0) partial_train_targets = np.concatenate([train_targets[:i * num_val_samples],train_targets[(i + 1) * num_val_samples:]], axis=0) model = Model() history = model.fit(partial_train_data, partial_train_targets, epochs=num_epochs, batch_size=batchsize, verbose=0) comparision1 = (model.predict(test_data))*100 Test_label = test_label*100 Comparision.append(comparision1) # r2_score(Test_label, comparision1, multioutput='raw_values') R2.append(r2_score(Test_label, comparision1, multioutput='raw_values')) # val_mse, val_mae = model.evaluate(val_data, val_targets, verbose=0) # mse_history = history.history['loss'] # mae_history = history.history['mean_absolute_error'] # val_mses.append(val_mse) # val_maes.append(val_mae) # mse_History.append(mse_history) # mae_History.append(mae_history) # return val_mses, val_maes, mse_History, mae_History return R2,Comparision R2_score, Predict_results = k_fold(5,train_data,train_label) # average_mae_history = [(np.mean([x[i] for x in mae_history]))*100 for i in range(50)] # average_mse_history = [(np.mean([x[i] for x in mse_history]))*100 for i in range(50)] # #model 3Dcnn # epochs = range(1,len(average_mse_history)+1) # plt.plot(epochs, average_mae_history,'c', label='mae' ) # plt.plot(epochs, average_mse_history,'m', label='mse' ) # plt.xlabel('Epochs') # plt.ylabel('Validation') # plt.show() data = np.load('/content/drive/My Drive/Newdata/shuffle_changedin4800.npy') label = np.load('/content/drive/My Drive/Newdata/shuffle_changedout4800.npy') data = np.reshape(data,(4800,30,116,116,1)) split_rate= 0.7 split = int(split_rate*4800) train_data = data[:split] train_label = label[:split] test_data = data[split:] test_label = label[split:] num_epochs = 50 batchsize = 6 model = Model() history = model.fit(train_data, train_label, epochs=num_epochs, batch_size=batchsize, verbose=0) # mse_history = history.history['loss'] # mae_history = history.history['mean_absolute_error'] comparision1 = (model.predict(test_data))*100 x_data = range(1080) Test_label = test_label*100 y_data_c1 = Test_label[:,0] y_data_c2 = Test_label[:,1] y_data_c3 = Test_label[:,2] y_predict1_c1 = comparision1[:,0] y_predict1_c2 = comparision1[:,1] y_predict1_c3 = comparision1[:,2] fig, (predict_c1, predict_c2, predict_c3) = plt.subplots(3,figsize=(18,6)) fig.suptitle('3 visual conditions_changed') # predict_c1.axes([0, 0.6, 1, 1]) # plt.subplots(figsize=(50, 10)) # plt.figure(figsize=(30,10) # fig, ax = plt.subplots(figsize=(20, 10)) # predict_c1.figure(figsize=(30,10) # bins = np.linspace(-1, 1, 21) #横坐标起始和结束值,分割成21份 # predict_c1.figure(figsize=(13,5)) #图像大小 # predict_c1.xticks(bins) #设置x轴 # predict_c1.xlim(-1, 1) #x轴开始和结束位置 # f = plt.figure(figsize=(20, 2)) # ax = f.add_subplot(111) # predict_c1.figure(figsize=(13,5)) # predict_c1.autoscale(enable=True, axis='x', tight=True)#plt.axis('tight') # predict_c1 predict_c1.plot(x_data, y_data_c1 , 'r', label='truth_c1') predict_c2.plot(x_data, y_data_c2 , 'r', label='truth_c2') predict_c3.plot(x_data, y_data_c3 , 'r', label='truth_c3') predict_c1.plot(x_data, y_predict1_c1 , 'g', label='predict1_c1') predict_c2.plot(x_data, y_predict1_c2 , 'g', label='predict1_c2') predict_c3.plot(x_data, y_predict1_c3 , 'g', label='predict1_c3') predict_c1.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) predict_c2.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) predict_c3.legend(bbox_to_anchor=(0.9, 1), loc='upper left', borderaxespad=0.) plt.savefig('/content/drive/My Drive/changed3600_50epo_0.7.png' ,dpi=1200) from sklearn.metrics import r2_score r2_score(Test_label, comparision1, multioutput='raw_values') print(R2_score) predict_c1.set_size_inches(18.5, 10.5, forward = True) # !cat ~/.keras/keras.json # from keras import backend # backend.set_image_data_format('channels_first') # print(backend.image_data_format()) # !cat ~/.keras/keras.json # # print(open("~/.keras/keras.json").read()) # content = {"epsilon": 1e-07, # "floatx": "float32", # "image_data_format": "channels_first", # "backend": "tensorflow"} # with open(" ~/.keras/keras.json",'w') as f: # f.write(str(content)) # # print(open("~/.keras/keras.json").read()) # tmp = data2[3,0,:,:] # # np.set_printoptions(threshold=np.inf) #打印全部内容 没有缩写 # print(tmp.shape) # # print(tmp[:,:]) # tmp=tmp.astype(np.float) # conver to float and plot # # %matplotlib qt # %matplotlib inline # plt.imshow(tmp,cmap='GnBu') # plt.savefig('/content/drive/My Drive/input_example.png' ,dpi=2400) # aaa = np.load('/content/drive/My Drive/output/M2.npy') # ppp = aaa[0] # print(ppp.shape) # ppp=ppp.astype(np.float) # conver to float and plot # # %matplotlib qt # %matplotlib inline # plt.imshow(ppp,cmap='gray') # plt.savefig('/content/drive/My Drive/output_example.png' ,dpi=2400) def SPLIT(data,split): data, label = shuffle(data, Label, random_state=20) train_data = data[:split] train_label = label[:split] test_data = data[split:] test_label = label[split:] return train_data, train_label, test_data, test_label train_data, train_label, test_data, test_label = SPLIT(data,72) # train_data1, train_label1, test_data1, test_label1 = SPLIT(data1,72) # train_data2, train_label2, test_data2, test_label2 = SPLIT(data2,72) def k_fold(k,train_data,train_targets): num_val_samples = len(train_data)//k val_mses = [] val_maes = [] mse_History = [] mae_History = [] num_epochs = 100 batchsize = 6 for i in range(k): print('processing fold #', i) val_data = train_data[i * num_val_samples: (i + 1) * num_val_samples] val_targets = train_targets[i * num_val_samples: (i + 1) * num_val_samples] partial_train_data = np.concatenate([train_data[:i * num_val_samples],train_data[(i + 1) * num_val_samples:]], axis=0) partial_train_targets = np.concatenate([train_targets[:i * num_val_samples],train_targets[(i + 1) * num_val_samples:]], axis=0) model = threeD_CNN1() history = model.fit(partial_train_data, partial_train_targets, epochs=num_epochs, batch_size=batchsize, verbose=0) val_mse, val_mae = model.evaluate(val_data, val_targets, verbose=0) mse_history = history.history['loss'] mae_history = history.history['mean_absolute_error'] val_mses.append(val_mse) val_maes.append(val_mae) mse_History.append(mse_history) mae_History.append(mae_history) return val_mses, val_maes, mse_History, mae_History val_mses1, val_maes1, mse_history1, mae_history1 = k_fold(4,train_data,train_label) average_mae_history1 = [(np.mean([x[i] for x in mae_history1]))*100 for i in range(100)] average_mse_history1= [(np.mean([x[i] for x in mse_history1]))*100 for i in range(100)] #model 3Dcnn epochs = range(1,len(average_mse_history1)+1) plt.plot(epochs, average_mae_history1,'c', label='mae' ) plt.plot(epochs, average_mse_history1,'m', label='mse' ) plt.xlabel('Epochs') plt.ylabel('Validation') plt.show() # batchsize = 6 # model1 = threeD_CNN() # history1 = model1.fit(train_data, train_label, epochs = 100, batch_size = batchsize, verbose=0) # model2 = threeD_CNN1() # history2 = model2.fit(train_data, train_label, epochs = 100, batch_size = batchsize,verbose=0) # model3 = threeD_CNN2() # history3 = model3.fit(train_data, train_label, epochs = 100,batch_size = batchsize, verbose=0) def k_fold(k,train_data,train_targets): num_val_samples = len(train_data)//k val_mses = [] val_maes = [] mse_History = [] mae_History = [] num_epochs = 100 batchsize = 6 for i in range(k): print('processing fold #', i) val_data = train_data[i * num_val_samples: (i + 1) * num_val_samples] val_targets = train_targets[i * num_val_samples: (i + 1) * num_val_samples] partial_train_data = np.concatenate([train_data[:i * num_val_samples],train_data[(i + 1) * num_val_samples:]], axis=0) partial_train_targets = np.concatenate([train_targets[:i * num_val_samples],train_targets[(i + 1) * num_val_samples:]], axis=0) model = threeD_CNN2() history = model.fit(partial_train_data, partial_train_targets, epochs=num_epochs, batch_size=batchsize, verbose=0) val_mse, val_mae = model.evaluate(val_data, val_targets, verbose=0) mse_history = history.history['loss'] mae_history = history.history['mean_absolute_error'] val_mses.append(val_mse) val_maes.append(val_mae) mse_History.append(mse_history) mae_History.append(mae_history) return val_mses, val_maes, mse_History, mae_History val_mses2, val_maes2, mse_history2, mae_history2 = k_fold(4,train_data,train_label) average_mae_history2 = [(np.mean([x[i] for x in mae_history2]))*100 for i in range(100)] average_mse_history2 = [(np.mean([x[i] for x in mse_history2]))*100 for i in range(100)] #model 3Dcnn epochs = range(1,len(average_mse_history2)+1) plt.plot(epochs, average_mae_history2,'c', label='mae' ) plt.plot(epochs, average_mse_history2,'m', label='mse' ) plt.xlabel('Epochs') plt.ylabel('Validation') plt.show() # batchsize = 6 # model1 = threeD_CNN() # history1 = model1.fit(train_data, train_label, epochs = 100, batch_size = batchsize, verbose=0) # model2 = threeD_CNN1() # history2 = model2.fit(train_data, train_label, epochs = 100, batch_size = batchsize,verbose=0) # model3 = threeD_CNN2() # history3 = model3.fit(train_data, train_label, epochs = 100,batch_size = batchsize, verbose=0) epochs = range(1,len(average_mse_history)+1) plt.plot(epochs, average_mae_history,'c', label='mae' ) plt.plot(epochs, average_mse_history,'c--', label='mse' ) plt.plot(epochs, average_mae_history1,'m', label='mae1' ) plt.plot(epochs, average_mse_history1,'m--', label='mse1' ) plt.plot(epochs, average_mae_history2,'r', label='mae2' ) plt.plot(epochs, average_mse_history2,'r--', label='mse2' ) plt.xlabel('Epochs') plt.ylabel('Average_performance') plt.legend() # plt.show() plt.savefig('/content/drive/My Drive/original_30_3models_comparison.png' ,dpi=1200) mse1 = history1.history['loss'] mae1 = history1.history['mean_absolute_error'] mse2 = history2.history['loss'] mae2 = history2.history['mean_absolute_error'] mse3 = history3.history['loss'] mae3 = history3.history['mean_absolute_error'] # val_loss = history.history['val_loss'] # acc = history.history['acc'] # val_acc=history.history['val_acc'] epochs = range(1,len(mse1) +1) plt.plot(epochs, mse1, 'c', label='mse1') plt.plot(epochs, mse2, 'm', label='mse2') plt.plot(epochs, mse3, 'r', label='mse3') plt.plot(epochs, mae1, 'c--', label='mae1') plt.plot(epochs, mae2, 'm--', label='mae2') plt.plot(epochs, mae3, 'r--', label='mae3') plt.title('loss') plt.xlabel('Epochs') plt.ylabel('Loss') plt.legend() plt.show() # plt.savefig('/content/drive/My Drive/original_30_3models_comparison.png' ,dpi=2400) comparision1 = (model1.predict(test_data))*100 comparision2 =(model2.predict(test_data))*100 comparision3 = (model3.predict(test_data))*100 print(comparision1[0,0]) print(comparision2) x_data = range(24) Test_label = test_label*100 y_data_c1 = Test_label[:,0] y_data_c2 = Test_label[:,1] y_data_c3 = Test_label[:,2] y_predict1_c1 = comparision1[:,0] y_predict1_c2 = comparision1[:,1] y_predict1_c3 = comparision1[:,2] y_predict2_c1 = comparision2[:,0] y_predict2_c2 = comparision2[:,1] y_predict2_c3 = comparision2[:,2] y_predict3_c1 = comparision3[:,0] y_predict3_c2 = comparision3[:,1] y_predict3_c3 = comparision3[:,2] fig, (predict_c1, predict_c2, predict_c3) = plt.subplots(3) fig.suptitle('3 visual conditions') predict_c1.plot(x_data, y_data_c1 , 'c', label='truth_c1') predict_c2.plot(x_data, y_data_c2 , 'r', label='truth_c2') predict_c3.plot(x_data, y_data_c3 , 'b', label='truth_c3') predict_c1.plot(x_data, y_predict1_c1 , 'c--', label='predict1_c1') predict_c2.plot(x_data, y_predict1_c2 , 'r--', label='predict1_c2') predict_c3.plot(x_data, y_predict1_c3 , 'b--', label='predict1_c3') predict_c1.plot(x_data, y_predict2_c1 , 'c:', label='predict2_c1') predict_c2.plot(x_data, y_predict2_c2 , 'r:', label='predict2_c2') predict_c3.plot(x_data, y_predict2_c3 , 'b:', label='predict2_c3') predict_c1.plot(x_data, y_predict3_c1 , 'c-.', label='predict2_c1') predict_c2.plot(x_data, y_predict3_c2 , 'r-.', label='predict2_c2') predict_c3.plot(x_data, y_predict3_c3 , 'b-.', label='predict2_c3') predict_c1.legend(bbox_to_anchor=(1.05, 1), loc='upper left', borderaxespad=0.) predict_c2.legend(bbox_to_anchor=(1.05, 1), loc='upper left', borderaxespad=0.) predict_c3.legend(bbox_to_anchor=(1.05, 1), loc='upper left', borderaxespad=0.) # plt.title('loss') # plt.xlabel('Epochs') # plt.ylabel('Loss') # plt.show() # fig.savefig('/content/drive/My Drive/ready_data_6C/matrix_input/1.png' ,bbox_inches='tight') print(comparision1.shape) y_data_c1 = test_label[:,0] print(y_data_c1.shape) def calculate_mse(predict): #对比三种视觉情况 每个model的mse y_data_c1 = test_label[:,0] y_data_c2 = test_label[:,1] y_data_c3 = test_label[:,2] y_predict1_c1 = predict[:,0] y_predict1_c2 = predict[:,1] y_predict1_c3 = predict[:,2] tmp1 = 0 tmp2 = 0 tmp3 = 0 for i in range(24): tmp1 = tmp1 + (y_data_c1[i]- y_predict1_c1[i])**2 tmp2 = tmp2 + (y_data_c2[i]- y_predict1_c2[i])**2 tmp3 = tmp3 + (y_data_c3[i]- y_predict1_c3[i])**2 c1 = tmp1 /24 c2 = tmp2 /24 c3 = tmp3 /24 # c1 = K.mean(K.square( y_data_c1- y_predict1_c1), axis=-1) # c2 = K.mean(K.square( y_data_c2- y_predict1_c2), axis=-1) # c3 = K.mean(K.square( y_data_c3- y_predict1_c3), axis=-1) return c1 , c2 ,c3 # from keras import backend as K Model1 = calculate_mse(comparision1) Model2 = calculate_mse(comparision2) print(Model1) print(Model2) <jupyter_output><empty_output>
no_license
/colab/Colab Notebooks/30到6_Matrix_run_file.ipynb
jackxuxu/Thesis
2
<jupyter_start><jupyter_text>## Challenge * The Inter-American Development Bank is asking the Kaggle community for help with income qualification for some of the world's poorest families. Are you up for the challenge? * Here's the backstory: Many social programs have a hard time making sure the right people are given enough aid. It’s especially tricky when a program focuses on the poorest segment of the population. The world’s poorest typically can’t provide the necessary income and expense records to prove that they qualify.## Road Map of Clean Data / Feature Engineering / Encoding / Remove Null Value 1. Load Data..... 1. Clean features... 1. Extracting features... 1. Encoding Data.... 1. Fill NA value..... 1. Prepared Model..... 1. Predict Value on test Dataset...... 1. Submit your result .......<jupyter_code>def dprint(*args, **kwargs): print("[{}] ".format(datetime.datetime.now().strftime("%Y-%m-%d %H:%M")) + \ " ".join(map(str,args)), **kwargs) id_name = 'Id' target_name = 'Target' # Load data train = pd.read_csv('../input/train.csv') test = pd.read_csv('../input/test.csv') train['is_test'] = 0 test['is_test'] = 1 df_all = pd.concat([train, test], axis=0) dprint('Clean features...') cols = ['dependency'] for c in tqdm(cols): x = df_all[c].values strs = [] for i, v in enumerate(x): try: val = float(v) except: strs.append(v) val = np.nan x[i] = val strs = np.unique(strs) for s in strs: df_all[c + '_' + s] = df_all[c].apply(lambda x: 1 if x == s else 0) df_all[c] = x df_all[c] = df_all[c].astype(float) dprint("Done.") dprint("Extracting features...") def extract_features(df): df['bedrooms_to_rooms'] = df['bedrooms']/df['rooms'] df['rent_to_rooms'] = df['v2a1']/df['rooms'] df['rent_to_bedrooms'] = df['v2a1']/df['bedrooms'] df['tamhog_to_rooms'] = df['tamhog']/df['rooms'] # tamhog - size of the household df['tamhog_to_bedrooms'] = df['tamhog']/df['bedrooms'] df['r4t3_to_tamhog'] = df['r4t3']/df['tamhog'] # r4t3 - Total persons in the household df['r4t3_to_rooms'] = df['r4t3']/df['rooms'] # r4t3 - Total persons in the household df['r4t3_to_bedrooms'] = df['r4t3']/df['bedrooms'] df['rent_to_r4t3'] = df['v2a1']/df['r4t3'] df['v2a1_to_r4t3'] = df['v2a1']/(df['r4t3'] - df['r4t1']) df['hhsize_to_rooms'] = df['hhsize']/df['rooms'] df['hhsize_to_bedrooms'] = df['hhsize']/df['bedrooms'] df['rent_to_hhsize'] = df['v2a1']/df['hhsize'] df['qmobilephone_to_r4t3'] = df['qmobilephone']/df['r4t3'] df['qmobilephone_to_v18q1'] = df['qmobilephone']/df['v18q1'] extract_features(train) extract_features(test) dprint("Done.") from sklearn.preprocessing import LabelEncoder def encode_data(df): yes_no_map = {'no': 0, 'yes': 1} df['dependency'] = df['dependency'].replace(yes_no_map).astype(np.float32) df['edjefe'] = df['edjefe'].replace(yes_no_map).astype(np.float32) df['edjefa'] = df['edjefa'].replace(yes_no_map).astype(np.float32) df['idhogar'] = LabelEncoder().fit_transform(df['idhogar']) dprint("Encoding Data....") encode_data(train) encode_data(test) dprint("Done...") def do_features(df): feats_div = [('children_fraction', 'r4t1', 'r4t3'), ('working_man_fraction', 'r4h2', 'r4t3'), ('all_man_fraction', 'r4h3', 'r4t3'), ('human_density', 'tamviv', 'rooms'), ('human_bed_density', 'tamviv', 'bedrooms'), ('rent_per_person', 'v2a1', 'r4t3'), ('rent_per_room', 'v2a1', 'rooms'), ('mobile_density', 'qmobilephone', 'r4t3'), ('tablet_density', 'v18q1', 'r4t3'), ('mobile_adult_density', 'qmobilephone', 'r4t2'), ('tablet_adult_density', 'v18q1', 'r4t2'), #('', '', ''), ] feats_sub = [('people_not_living', 'tamhog', 'tamviv'), ('people_weird_stat', 'tamhog', 'r4t3')] for f_new, f1, f2 in feats_div: df['fe_' + f_new] = (df[f1] / df[f2]).astype(np.float32) for f_new, f1, f2 in feats_sub: df['fe_' + f_new] = (df[f1] - df[f2]).astype(np.float32) # aggregation rules over household aggs_num = {'age': ['min', 'max', 'mean'], 'escolari': ['min', 'max', 'mean'] } aggs_cat = {'dis': ['mean']} for s_ in ['estadocivil', 'parentesco', 'instlevel']: for f_ in [f_ for f_ in df.columns if f_.startswith(s_)]: aggs_cat[f_] = ['mean', 'count'] # aggregation over household for name_, df_ in [('18', df.query('age >= 18'))]: df_agg = df_.groupby('idhogar').agg({**aggs_num, **aggs_cat}).astype(np.float32) df_agg.columns = pd.Index(['agg' + name_ + '_' + e[0] + "_" + e[1].upper() for e in df_agg.columns.tolist()]) df = df.join(df_agg, how='left', on='idhogar') del df_agg # do something advanced above... # Drop SQB variables, as they are just squres of other vars df.drop([f_ for f_ in df.columns if f_.startswith('SQB') or f_ == 'agesq'], axis=1, inplace=True) # Drop id's df.drop(['Id', 'idhogar'], axis=1, inplace=True) # Drop repeated columns df.drop(['hhsize', 'female', 'area2'], axis=1, inplace=True) return df dprint("Do_feature Engineering....") train = do_features(train) test = do_features(test) dprint("Done....") dprint("Fill Na value....") train = train.fillna(0) test = test.fillna(0) dprint("Done....") train.shape,test.shape cols_to_drop = [ id_name, target_name, ] X = train.drop(cols_to_drop, axis=1, errors='ignore') y = train[target_name].values y = pd.get_dummies(y).values from sklearn.model_selection import train_test_split X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, random_state=42, stratify=y) X_train.shape,y_train.shape,X_test.shape,y_test.shape import keras from keras.models import Sequential from keras.layers import Dense, Dropout, Flatten from keras.layers import Conv2D, MaxPooling2D from keras.layers.normalization import BatchNormalization from keras.optimizers import Adam from keras import regularizers input_dim = len(X_train.columns) input_dim input_dim = len(X_train.columns) batch_size=32 model = Sequential() # first input layer with first hidden layer in a single statement model.add( Dense(120, input_shape=(input_dim,), activation='tanh') ) # 10 is the size(no. of neurons) of first hidden layer, 4 is the no. of features in the input layer # input_shape=(4,) can also be written as input_dim=4 # second hiden layer model.add(Dense(64,activation='relu')) # 8 = no. of neurons in second hidden layer # third hiden layer model.add(Dense(32,activation='relu')) # 6 = no. of neurons in third hidden layer # ouput layer model.add(Dense(4,activation='softmax')) # 3 = no. of neurons in output layer as three categories of labels are there # compile method receives three arguments: "an optimizer", "a loss function" and "a list of metrics" model.compile(Adam(lr=0.02),'mse', ['accuracy']) # we use "binary_crossentropy" for binary classification problems and # "categorical_crossentropy" for multiclass classification problems # the compile statement can also be written as:- # model.compile(optimizer=Adam(lr=0.04), loss='categorical_crossentropy',metrics=['accuracy']) # we can give more than one metrics like ['accuracy', 'mae', 'mape'] model.summary() model.fit(X_train, y_train, steps_per_epoch=850 ,epochs = 10) scores = model.evaluate(X_test, y_test) print("\n%s: %.2f%%" % (model.metrics_names[1], scores[1]*100)) predictions = model.predict_classes(test) list(set(predictions)) sub = pd.read_csv("../input/sample_submission.csv") sub['Target'] = predictions sub.to_csv("keras.csv", index= False)<jupyter_output><empty_output>
no_license
/datasets/costa-rican-household-poverty-prediction/kernels/0.12---nikitpatel---keras-deeplearning-classification.ipynb
mindis/GDS
1
<jupyter_start><jupyter_text># 1) Derive the descriptive statistics of the data and discuss the points you find remarkable.<jupyter_code>usEducation.head() fill_list = ["enroll", "total_revenue", "federal_revenue", "state_revenue", "federal_revenue", "total_expenditure", "instruction_expenditure", "support_services_expenditure", "other_expenditure", "capital_outlay_expenditure", "grades_pk_g", "grades_kg_g", "grades_4_g", "grades_8_g", "grades_12_g", "grades_all_g", "avg_math_4_score", "avg_math_8_score",'avg_reading_4_score','avg_reading_4_score','avg_reading_8_score'] for col in fill_list: usEducation.loc[:, col] = usEducation.loc[:, col].fillna(usEducation.loc[:, col].mean()) usEducation.isnull().sum()/usEducation.isnull().count() years = usEducation["year"].unique() for col in fill_list: for year in years: usEducation.loc[usEducation["year"] == year, col] = usEducation.loc[usEducation["year"] == year, col].fillna( usEducation[usEducation["year"] == year][col].mean()) usEducation.describe()<jupyter_output><empty_output><jupyter_text>1) Total Revenue and Federal Revenue are close in numbers. Meaning most revenue goes to federal. 2)Total Expenditure is greater than Total Revenue when we compare the mean, median, and max. More money was spent than what we received as revenue. 3) Instruction Expenditure is about half of Total Revenue. 4) On average the math scores for each grades are higher than average reading score. Std for average math is about the same # 2) Choose a state (e.g. California) and draw a line graph of its total revenues and total expenditures along the years.<jupyter_code>plt.plot(usEducation.loc[usEducation.state=='MINNESOTA', 'year'], usEducation.loc[usEducation.state=='MINNESOTA', 'total_revenue'], label ='Total Revenue') plt.title("Minnesota's Total Revenue VS Total Expenditure") plt.plot(usEducation.loc[usEducation.state=='MINNESOTA', 'year'], usEducation.loc[usEducation.state=='MINNESOTA', 'total_expenditure'], label ='Total Expenditure') plt.legend() plt.show()<jupyter_output><empty_output><jupyter_text># 3) In your chosen state, which of the lessons are the students more successful, math or reading?<jupyter_code>plt.title('The difference in reading and math scores for 4') plt.plot(usEducation.loc[usEducation.state=='MINNESOTA', 'year'], usEducation.loc[usEducation.state=='MINNESOTA', 'avg_math_4_score'], label ='Avg Math Score') plt.plot(usEducation.loc[usEducation.state=='MINNESOTA', 'year'], usEducation.loc[usEducation.state == 'MINNESOTA', 'avg_reading_4_score'], label ='Avg Reading Score') plt.legend() plt.show() plt.title('The difference in reading and math scores for 8') plt.plot(usEducation.loc[usEducation.state=='MINNESOTA', 'year'], usEducation.loc[usEducation.state=='MINNESOTA', 'avg_math_4_score'], label = 'Avg Math Score') plt.plot(usEducation.loc[usEducation.state=='MINNESOTA', 'year'], usEducation.loc[usEducation.state =='MINNESOTA', 'avg_reading_8_score'], label = 'Avg Reading Score') plt.show() <jupyter_output><empty_output><jupyter_text>It seems students are more successful in math than reading.# 4) What are the distributions of the math and reading scores in the sample?<jupyter_code>plt.figure(figsize=(20,10)) plt.subplot(2,2,1) plt.hist(usEducation.avg_math_4_score.dropna()) plt.title("histogram of {}".format("avg_math_4_score")) plt.subplot(2,2,2) plt.hist(usEducation.avg_reading_4_score.dropna()) plt.title("histogram of {}".format("avg_reading_4_score")) plt.subplot(2,2,3) plt.hist(usEducation.avg_math_8_score.dropna()) plt.title("histogram of {}".format("avg_math_8_score")) plt.subplot(2,2,4) plt.hist(usEducation.avg_reading_8_score.dropna()) plt.title("histogram of {}".format("avg_reading_8_score")) plt.show()<jupyter_output><empty_output><jupyter_text>Distributions are not normal, all are right skweed.# 4) 5. Notice that there are too many missing values for math and reading scores. Fill out the missing values using mean, median and linear interpolation. Then compare the effects of these techniques on the distributions of the score variables.¶<jupyter_code>postgres_user = 'dsbc_student' postgres_pw = '7*.8G9QH21' postgres_host = '142.93.121.174' postgres_port = '5432' postgres_db = 'useducation' engine = create_engine('postgresql://{}:{}@{}:{}/{}'.format( postgres_user, postgres_pw, postgres_host, postgres_port, postgres_db)) usEducation = pd.read_sql_query('select * from useducation',con=engine) engine.dispose() usEducation.columns = usEducation.columns.str.lower() plt.figure(figsize=(20,20)) plt.subplot(4,4,1) plt.hist(usEducation.avg_math_4_score.dropna()) plt.title("histogram of {} (original)".format("avg_math_4_score")) plt.subplot(4,4,2) plt.hist(usEducation.avg_math_4_score.interpolate()) plt.title("histogram of {} (interpolated)".format("avg_reading_4_score")) plt.subplot(4,4,3) plt.hist(usEducation.avg_math_4_score.fillna(usEducation.avg_math_4_score.median())) plt.title("histogram of {} (filled with median)".format("avg_math_4_score")) plt.subplot(4,4,4) plt.hist(usEducation.avg_math_4_score.fillna(usEducation.avg_math_4_score.mean())) plt.title("histogram of {} (filled with mean)".format("avg_math_4_score")) plt.subplot(4,4,5) plt.hist(usEducation.avg_reading_4_score.dropna()) plt.title("histogram of {} (original)".format("avg_reading_4_score")) plt.subplot(4,4,6) plt.hist(usEducation.avg_reading_4_score.interpolate()) plt.title("histogram of {} (interpolated)".format("avg_reading_4_score")) plt.subplot(4,4,7) plt.hist(usEducation.avg_reading_4_score.fillna(usEducation.avg_reading_4_score.median())) plt.title("histogram of {} (filled with median)".format("avg_reading_4_score")) plt.subplot(4,4,8) plt.hist(usEducation.avg_reading_4_score.fillna(usEducation.avg_reading_4_score.mean())) plt.title("histogram of {} (filled with mean)".format("avg_reading_4_score")) plt.subplot(4,4,9) plt.hist(usEducation.avg_math_8_score.dropna()) plt.title("histogram of {} (original)".format("avg_math_8_score")) plt.subplot(4,4,10) plt.hist(usEducation.avg_math_8_score.interpolate()) plt.title("histogram of {} (interpolated)".format("avg_math_8_score")) plt.subplot(4,4,11) plt.hist(usEducation.avg_math_8_score.fillna(usEducation.avg_math_8_score.median())) plt.title("histogram of {} (filled with median)".format("avg_math_8_score")) plt.subplot(4,4,12) plt.hist(usEducation.avg_math_8_score.fillna(usEducation.avg_math_8_score.mean())) plt.title("histogram of {} (filled with mean)".format("avg_math_8_score")) plt.subplot(4,4,13) plt.hist(usEducation.avg_reading_8_score.dropna()) plt.title("histogram of {} (original)".format("avg_reading_8_score")) plt.subplot(4,4,14) plt.hist(usEducation.avg_reading_8_score.interpolate().dropna()) plt.title("histogram of {} (interpolated)".format("avg_reading_8_score")) plt.subplot(4,4,15) plt.hist(usEducation.avg_reading_8_score.fillna(usEducation.avg_reading_8_score.median())) plt.title("histogram of {} (filled with median)".format("avg_reading_8_score")) plt.subplot(4,4,16) plt.hist(usEducation.avg_reading_8_score.fillna(usEducation.avg_reading_8_score.mean())) plt.title("histogram of {} (filled with mean)".format("avg_reading_8_score")) plt.tight_layout() plt.show()<jupyter_output><empty_output>
no_license
/explorationPhase1.ipynb
mysticalScientist23/HR-Employee-Attrition
5
<jupyter_start><jupyter_text>![image.png](attachment:image.png)<jupyter_code>R = 6371.0 #* 1000 #in m atan = math.atan sin = math.sin cos = math.cos pi = math.pi answer = 4 distancelist = [] for i in range(0,answer): if i+1 == (answer): lon1 = latlist[answer-1] lon2 = latlist[0] lat1 = lonlist[answer-1] lat2 = lonlist[0] dlon = lon2 - lon1 dlat = lat2 - lat1 #Haversine formula a = math.sin(dlat / 2)**2 + math.cos(lat1) * math.cos(lat2) * math.sin(dlon / 2)**2 c = 2 * math.atan2(math.sqrt(a), math.sqrt(1 - a)) distance = R * c print('The distance between coordinate %s and coordinate %s is %s' %(i, 0, round(distance,4) )) else: lon1 = latlist[i] lon2 = latlist[i+1] lat1 = lonlist[i] lat2 = lonlist[i+1] dlon = lon2 - lon1 dlat = lat2 - lat1 #Haversine formula a = math.sin(dlat / 2)**2 + math.cos(lat1) * math.cos(lat2) * math.sin(dlon / 2)**2 c = 2 * math.atan2(math.sqrt(a), math.sqrt(1 - a)) distance = R * c print('The distance between coordinate %s and coordinate %s is %s ' %(i, i+1, round(distance,4) )) distancelist.append(distance) origindist = [] for i in range(1,answer): lon1 = latlist[0] lon2 = latlist[i] lat1 = lonlist[0] lat2 = lonlist[i] dlon = lon2 - lon1 dlat = lat2 - lat1 #Haversine formula a = math.sin(dlat / 2)**2 + math.cos(lat1) * math.cos(lat2) * math.sin(dlon / 2)**2 c = 2 * math.atan2(math.sqrt(a), math.sqrt(1 - a)) distance = R * c origindist.append(distance) print(lonlist) print(latlist) plt.plot(lonlist[0],latlist[0], 'bo') plt.plot(lonlist[1],latlist[1], 'go') plt.plot(lonlist[2],latlist[2], 'yo') plt.plot(lonlist[3],latlist[3], 'ro') def slope(x1,y1,x2,y2): m = round( ((y2-y1)/(x2-x1)) ,6) return m mlist = [] for i in range(0, answer): if i+1 == answer: m = slope(lonlist[0],latlist[0],lonlist[answer-1],latlist[answer-1]) else: m= slope(lonlist[i],latlist[i],lonlist[i+1],latlist[i+1]) mlist.append(m) originslope = [] for i in range(1, answer): m= slope(lonlist[0],latlist[0],lonlist[i],latlist[i]) print(m) originslope.append(m) mlist #assuming the first lat and lon are 0, 0 # rise/run use slope, not angles def coordinates_fuc(i,x,y,m,d): print(i) print(m) angle = atan(m) print(angle) if i < 2: if angle < 0: angle = pi + angle else: if angle < 0: angle = 2*pi + angle x = d*cos(angle) y = d*sin(angle) print(x) print(y) print("\n") return x,y xlist = [0] ylist = [0] for i in range (0, answer-1): x,y = coordinates_fuc(i,xlist[0],ylist[0],originslope[i],origindist[i]) xlist.append(x) ylist.append(y) print(xlist) print(ylist) plt.plot(xlist[0],ylist[0], 'bo') plt.plot(xlist[1],ylist[1], 'go') plt.plot(xlist[2],ylist[2], 'yo') plt.plot(xlist[3],ylist[3], 'ro') xmin = min(xlist) ymin = min(ylist) for i in range(0, answer): xlist[i] = xlist[i] - xmin ylist[i] = ylist[i] - ymin plt.plot(xlist[0],ylist[0], 'bo') plt.plot(xlist[1],ylist[1], 'go') plt.plot(xlist[2],ylist[2], 'yo') plt.plot(xlist[3],ylist[3], 'ro') def cvalue(x1,y1,x2,y2,m): c1 = round((-m*x1 + y1),7) c2 = round((-m*x2 + y2),7) print(c1) print(c2) if c1 != c2: print('Error') return c1 clist = [] for i in range(0, answer): if i+1 == answer: c = cvalue(xlist[0],ylist[0],xlist[answer-1],ylist[answer-1],mlist[answer-1]) else: c = cvalue(xlist[i],ylist[i],xlist[i+1],ylist[i+1],mlist[i]) clist.append(c) print(clist) xmax = max(xlist) ymax = max(ylist) #SPLIT PROBLEM IN TWO xlist.append( (xlist[1] + xlist[0])/2 ) ylist.append( (ylist[1] + ylist[0])/2 ) xlist.append( (xlist[2] + xlist[3])/2 ) ylist.append( (ylist[2] + ylist[3])/2 ) #SPLIT PROBLEM IN TWO xlist.append( (xlist[0] + xlist[3])/2 ) ylist.append( (ylist[0] + ylist[3])/2 ) xlist.append( (xlist[1] + xlist[2])/2 ) ylist.append( (ylist[1] + ylist[2])/2 ) #SPLIT PROBLEM IN TWO xlist.append( (xlist[4] + xlist[5])/2 ) ylist.append( (ylist[4] + ylist[5])/2 ) xlist.append( (xlist[6] + xlist[7])/2 ) ylist.append( (ylist[6] + ylist[7])/2 ) print(xlist) print(ylist) m= slope(xlist[4],ylist[4],xlist[5],ylist[5]) mlist.append(m) c = cvalue(xlist[4],ylist[4],xlist[5],ylist[5],mlist[4]) clist.append(c) m= slope(xlist[6],ylist[6],xlist[7],ylist[7]) mlist.append(m) c = cvalue(xlist[6],ylist[6],xlist[7],ylist[7],mlist[5]) clist.append(c) print(clist) print(mlist) plt.plot(xlist[0],ylist[0], 'bo') plt.plot(xlist[1],ylist[1], 'go') plt.plot(xlist[2],ylist[2], 'yo') plt.plot(xlist[3],ylist[3], 'ro') plt.plot(xlist[4],ylist[4], 'mo') plt.plot(xlist[5],ylist[5], 'co') plt.plot(xlist[6],ylist[6], 'mo') plt.plot(xlist[7],ylist[7], 'co') plt.plot(xlist[8],ylist[8], 'ro') plt.plot(xlist[9],ylist[9], 'bo')<jupyter_output><empty_output><jupyter_text>### The pyomo code is taken from Souridi (2020) (A. Soroudi, “Lecture 14 - Heuristic Methods,” University College Dublin, Dublin, 2020.)<jupyter_code>#Section A xinitiallist = [] yinitiallist = [] for j in range(0,2000): #more the better, x = random.randint(0, ceil(xmax)) # random generation y = random.randint(0, ceil(ymax)) if (y<=mlist[0]*x+clist[0]) & (y<=mlist[1]*x+clist[1]) & (y<=mlist[5]*x+clist[5]) & (y>=mlist[4]*x+clist[4]): xinitiallist.append(x) yinitiallist.append(y) print(len(xinitiallist)) model = AbstractModel() model.N = Param(mutable=True) model.i = RangeSet(1, model.N) model.j = Set(initialize=model.i) # model.L=Param(initialize=100000,mutable=True) # def initval(model,i): # return random.uniform(0,model.L) model.L=Param(initialize=100000,mutable=True) def initvalx(model,i): k = randrange(0,len(xinitiallist)) return xinitiallist[k] def initvaly(model,i): k = randrange(0,len(yinitiallist)) return yinitiallist[k] model.x = Var(model.i , bounds=(0,xmax+1), within=NonNegativeReals, initialize=initvalx) model.y = Var(model.i ,bounds=(0,ymax+1) , within=NonNegativeReals, initialize=initvaly) model.r = Var(bounds=(0,xmax), within=NonNegativeReals) def C1_rule(model,i,j): if i<j: return (model.x[i]-model.x[j])**2+(model.y[i]-model.y[j])**2 >= model.r**2 else: return Constraint.Skip model.C = Constraint(model.i,model.j, rule=C1_rule) def C2rule(model,i): return model.y[i] <= mlist[0]*model.x[i] + clist[0] model.C2 = Constraint(model.i,rule=C2rule ) def C3rule(model,i): return model.y[i] <= mlist[1]*model.x[i] + clist[1] model.C3 = Constraint(model.i, rule=C3rule ) def C4rule(model,i): return model.y[i] <= mlist[5]*model.x[i] + clist[5] model.C4 = Constraint(model.i, rule=C4rule ) def C5rule(model,i): return model.y[i] >= mlist[4]*model.x[i] + clist[4] model.C5 = Constraint(model.i, rule=C5rule ) # y=ax+b y=1x+3 model.obj = Objective(expr=model.r, sense=maximize) #opt = SolverFactory('ipopt') # Section A plt.plot(xlist[1],ylist[1], 'go') plt.plot(xlist[4],ylist[4], 'mo') plt.plot(xlist[7],ylist[7], 'co') plt.plot(xlist[8],ylist[8], 'bo') print("Analysing wind turbine %s" %wt) print("%s has a radius of %s" %(wt,radius)) print('Min distance should be %s km' %(round(min_distance,3))) print('\n') model.N=38 # 38 = 1.812 (exceeded) instance = model.create_instance() #results = opt.solve(instance) # solves and updates instance results =SolverFactory('multistart').solve(instance) X=[value(instance.x[i]) for i in instance.i] Y=[value(instance.y[i]) for i in instance.i] plt.figure(figsize =(20,10)) plt.scatter( X,Y,s=25,color='blue') plt.xlabel('X',size=20) plt.ylabel('Y',size=20) plt.title('Wind farms for N=%s' %( (str(value(instance.N)))), size=20) print('Min distance is ',round(value(instance.r),3)) plt.show() #Section B xinitiallist = [] yinitiallist = [] for j in range(0,2000): #more the better, x = random.randint(0, ceil(xmax)) # random generation y = random.randint(0, ceil(ymax)) if (y>=mlist[5]*x+clist[5]) & (y<=mlist[1]*x+clist[1]) & (y<=mlist[2]*x+clist[2]) & (y>=mlist[4]*x+clist[4]): xinitiallist.append(x) yinitiallist.append(y) print(len(xinitiallist)) model = AbstractModel() model.N = Param(mutable=True) model.i = RangeSet(1, model.N) model.j = Set(initialize=model.i) # model.L=Param(initialize=100000,mutable=True) # def initval(model,i): # return random.uniform(0,model.L) model.L=Param(initialize=100000,mutable=True) def initvalx(model,i): k = randrange(0,len(xinitiallist)) return xinitiallist[k] def initvaly(model,i): k = randrange(0,len(yinitiallist)) return yinitiallist[k] model.x = Var(model.i , bounds=(0,xmax+1), within=NonNegativeReals, initialize=initvalx) model.y = Var(model.i ,bounds=(0,ymax+1) , within=NonNegativeReals, initialize=initvaly) model.r = Var(bounds=(0,xmax), within=NonNegativeReals) def C1_rule(model,i,j): if i<j: return (model.x[i]-model.x[j])**2+(model.y[i]-model.y[j])**2 >= model.r**2 else: return Constraint.Skip model.C = Constraint(model.i,model.j, rule=C1_rule) def C2rule(model,i): return model.y[i] >= mlist[5]*model.x[i] + clist[5] model.C2 = Constraint(model.i,rule=C2rule ) def C3rule(model,i): return model.y[i] <= mlist[1]*model.x[i] + clist[1] model.C3 = Constraint(model.i, rule=C3rule ) def C4rule(model,i): return model.y[i] <= mlist[2]*model.x[i] + clist[2] model.C4 = Constraint(model.i, rule=C4rule ) def C5rule(model,i): return model.y[i] >= mlist[4]*model.x[i] + clist[4] model.C5 = Constraint(model.i, rule=C5rule ) # y=ax+b y=1x+3 model.obj = Objective(expr=model.r, sense=maximize) #opt = SolverFactory('ipopt') # Section B plt.plot(xlist[2],ylist[2], 'yo') plt.plot(xlist[5],ylist[5], 'co') plt.plot(xlist[7],ylist[7], 'co') plt.plot(xlist[8],ylist[8], 'bo') print("Analysing wind turbine %s" %wt) print("%s has a radius of %s" %(wt,radius)) print('Min distance should be %s metres' %(round(min_distance,3))) print('\n') model.N= 35 # 35 = 1.641 instance = model.create_instance() #results = opt.solve(instance) # solves and updates instance results =SolverFactory('multistart').solve(instance) X=[value(instance.x[i]) for i in instance.i] Y=[value(instance.y[i]) for i in instance.i] plt.figure(figsize =(20,10)) plt.scatter( X,Y,s=25,color='blue') plt.xlabel('X',size=20) plt.ylabel('Y',size=20) plt.title('Wind farms for N=%s' %( (str(value(instance.N)))), size=20) print('Min distance is ',round(value(instance.r),3)) plt.show() #Section C xinitiallist = [] yinitiallist = [] for j in range(0,2000): #more the better, x = random.randint(0, ceil(xmax)) # random generation y = random.randint(0, ceil(ymax)) if (y>=mlist[5]*x+clist[5]) & (y<=mlist[4]*x+clist[4]) & (y<=mlist[2]*x+clist[2]) & (y>=mlist[3]*x+clist[3]): xinitiallist.append(x) yinitiallist.append(y) print(len(xinitiallist)) model = AbstractModel() model.N = Param(mutable=True) model.i = RangeSet(1, model.N) model.j = Set(initialize=model.i) # model.L=Param(initialize=100000,mutable=True) # def initval(model,i): # return random.uniform(0,model.L) model.L=Param(initialize=100000,mutable=True) def initvalx(model,i): k = randrange(0,len(xinitiallist)) return xinitiallist[k] def initvaly(model,i): k = randrange(0,len(yinitiallist)) return yinitiallist[k] model.x = Var(model.i , bounds=(0,xmax+1), within=NonNegativeReals, initialize=initvalx) model.y = Var(model.i ,bounds=(0,ymax+1) , within=NonNegativeReals, initialize=initvaly) model.r = Var(bounds=(0,xmax), within=NonNegativeReals) def C1_rule(model,i,j): if i<j: return (model.x[i]-model.x[j])**2+(model.y[i]-model.y[j])**2 >= model.r**2 else: return Constraint.Skip model.C = Constraint(model.i,model.j, rule=C1_rule) def C2rule(model,i): return model.y[i] >= mlist[5]*model.x[i] + clist[5] model.C2 = Constraint(model.i,rule=C2rule ) def C3rule(model,i): return model.y[i] <= mlist[4]*model.x[i] + clist[4] model.C3 = Constraint(model.i, rule=C3rule ) def C4rule(model,i): return model.y[i] <= mlist[2]*model.x[i] + clist[2] model.C4 = Constraint(model.i, rule=C4rule ) def C5rule(model,i): return model.y[i] >= mlist[3]*model.x[i] + clist[3] model.C5 = Constraint(model.i, rule=C5rule ) # y=ax+b y=1x+3 model.obj = Objective(expr=model.r, sense=maximize) #opt = SolverFactory('ipopt') # Section C plt.plot(xlist[3],ylist[3], 'ro') plt.plot(xlist[5],ylist[5], 'co') plt.plot(xlist[6],ylist[6], 'mo') plt.plot(xlist[8],ylist[8], 'bo') print("Analysing wind turbine %s" %wt) print("%s has a radius of %s" %(wt,radius)) print('Min distance should be %s metres' %(round(min_distance,3))) print('\n') model.N= 34 #36 = 1.913 (exceeded) #34 = 1.979 instance = model.create_instance() #results = opt.solve(instance) # solves and updates instance results =SolverFactory('multistart').solve(instance) X=[value(instance.x[i]) for i in instance.i] Y=[value(instance.y[i]) for i in instance.i] plt.figure(figsize =(20,10)) plt.scatter( X,Y,s=25,color='blue') plt.xlabel('X',size=20) plt.ylabel('Y',size=20) plt.title('Wind farms for N=%s' %( (str(value(instance.N)))), size=20) print('Min distance is ',round(value(instance.r),3)) plt.show() #Section D xinitiallist = [] yinitiallist = [] for j in range(0,20000): #more the better, x = random.randint(0, ceil(xmax)) # random generation y = random.randint(0, ceil(ymax)) if (y<=mlist[0]*x+clist[0]) & (y<=mlist[4]*x+clist[4]) & (y<=mlist[5]*x+clist[5]) & (y>=mlist[3]*x+clist[3]): xinitiallist.append(x) yinitiallist.append(y) print(len(xinitiallist)) model = AbstractModel() model.N = Param(mutable=True) model.i = RangeSet(1, model.N) model.j = Set(initialize=model.i) # model.L=Param(initialize=100000,mutable=True) # def initval(model,i): # return random.uniform(0,model.L) model.L=Param(initialize=100000,mutable=True) def initvalx(model,i): k = randrange(0,len(xinitiallist)) return xinitiallist[k] def initvaly(model,i): k = randrange(0,len(yinitiallist)) return yinitiallist[k] model.x = Var(model.i , bounds=(0,xmax+1), within=NonNegativeReals, initialize=initvalx) model.y = Var(model.i ,bounds=(0,ymax+1) , within=NonNegativeReals, initialize=initvaly) model.r = Var(bounds=(0,xmax), within=NonNegativeReals) def C1_rule(model,i,j): if i<j: return (model.x[i]-model.x[j])**2+(model.y[i]-model.y[j])**2 >= model.r**2 else: return Constraint.Skip model.C = Constraint(model.i,model.j, rule=C1_rule) def C2rule(model,i): return model.y[i] <= mlist[0]*model.x[i] + clist[0] model.C2 = Constraint(model.i,rule=C2rule ) def C3rule(model,i): return model.y[i] <= mlist[4]*model.x[i] + clist[4] model.C3 = Constraint(model.i, rule=C3rule ) def C4rule(model,i): return model.y[i] <= mlist[5]*model.x[i] + clist[5] model.C4 = Constraint(model.i, rule=C4rule ) def C5rule(model,i): return model.y[i] >= mlist[3]*model.x[i] + clist[3] model.C5 = Constraint(model.i, rule=C5rule ) # y=ax+b y=1x+3 model.obj = Objective(expr=model.r, sense=maximize) #opt = SolverFactory('ipopt') # Section D plt.plot(xlist[0],ylist[0], 'bo') plt.plot(xlist[4],ylist[4], 'mo') plt.plot(xlist[6],ylist[6], 'mo') plt.plot(xlist[8],ylist[8], 'bo') print("Analysing wind turbine %s" %wt) print("%s has a radius of %s" %(wt,radius)) print('Min distance should be %s metres' %(round(min_distance,3))) print('\n') model.N= 40 #40 = 2.059 #41 = 2.01, 48 infeasible instance = model.create_instance() #results = opt.solve(instance) # solves and updates instance results =SolverFactory('multistart').solve(instance) X=[value(instance.x[i]) for i in instance.i] Y=[value(instance.y[i]) for i in instance.i] plt.figure(figsize =(20,10)) plt.scatter( X,Y,s=25,color='blue') plt.xlabel('X',size=20) plt.ylabel('Y',size=20) plt.title('Wind farms for N=%s' %( (str(value(instance.N)))), size=20) print('Min distance is ',round(value(instance.r),3)) plt.show() #instance.C.pprint()<jupyter_output><empty_output>
no_license
/Section 2 - Optimising Wind Farm Spacing Rosslare Harbour to Kilmore Quay.ipynb
16431496/optimisation_code
2
<jupyter_start><jupyter_text>## Task 4: **Healthcare Sector Analysis** Discovering facts from data in healthcare sector<jupyter_code>import os import re import sys import json import nltk import pandas as pd import numpy as np from scipy.stats import norm, ttest_ind from collections import defaultdict import matplotlib.pyplot as plt from sklearn.linear_model import LinearRegression from sklearn.model_selection import train_test_split from sklearn.metrics import mean_squared_error, r2_score from sklearn.preprocessing import PolynomialFeatures from sklearn.svm import SVR # from google.colab import drive # drive.mount('/content/drive') def read_data(root_dir, data_folder, csv_file): if data_folder is not None: print("\nReading data from " + csv_file) file_dir = os.path.join(root_dir, data_folder, csv_file) return_df = pd.read_csv(file_dir) return return_df else: print("\nReading data from " + csv_file) file_dir = os.path.join(root_dir, csv_file) return_df = pd.read_csv(file_dir) return return_df def merge_dataframe(dst_dataframe, new_part): no_nan_new_part = new_part.copy().fillna('') if dst_dataframe is None: dst_dataframe = no_nan_new_part.copy() return dst_dataframe else: dst_dataframe = pd.concat([dst_dataframe, no_nan_new_part]) return dst_dataframe ''' Basic environ params ''' root_dir = "\\".join(os.path.abspath('').split('\\')[:-1]) # root_dir = '/content/drive/My Drive/Penn Inequality Project' task_folder = "task4" ''' Main adjustable params ''' # Section of codes, which are listed in every cell company_analyzed = False reference_level_analyzed = False incumbent_point_analyzed = False skill_level_analyzed = False min_surveying_years = 3 min_level_counts = 5 max_level_counts = 20 print("Read csv file from path " + root_dir) data_df = read_data(root_dir, None, 'reclassified_all_data.csv') data_df.columns<jupyter_output><empty_output><jupyter_text>## Company Analysis### 1. Raw analysis of all companies in healthcare sector<jupyter_code>if not company_analyzed: # Mean data_df = data_df[['KF_ID', 'CalendarYear', 'SectorName', 'Base Salary', 'RegionName', 'OwnershipTypeDesc', 'NumOfEmpDesc', 'ReferenceLevelNum', 'IncumbentPointCount', 'JobName', 'RegionName']] healthcare_df = data_df[data_df['SectorName']=="healthcare"] healthcare_df = healthcare_df[healthcare_df['ReferenceLevelNum'] != 99] print("A total of " + str(healthcare_df.shape[0]) + " data pieces are related with healthcare industry\n") healthcare_mean_salary = healthcare_df['Base Salary'].mean() print("The mean salary of the healthcare industry is " + str(healthcare_mean_salary) + "\n") annual_mean = [] for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] company_year_mean = year_healthcare_df[['KF_ID', 'Base Salary']].groupby(by = ['KF_ID']).mean() annual_mean_salary = company_year_mean['Base Salary'].mean() annual_mean.append(annual_mean_salary) fig1 = plt.figure() plt.title("The annual mean salary of healthcare industry") plt.ylabel("Mean salary value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), annual_mean) plt.show() # Median healthcare_median_salary = healthcare_df['Base Salary'].median() print("\nThe median salary of the healthcare industry is " + str(healthcare_median_salary) + "\n") annual_median = [] for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] company_year_median = year_healthcare_df[['KF_ID', 'Base Salary']].groupby(by = ['KF_ID']).median() annual_median_salary = company_year_median['Base Salary'].median() annual_median.append(annual_median_salary) fig2 = plt.figure() plt.title("The annual median salary of healthcare industry") plt.ylabel("Median salary value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), annual_median) plt.show() print("\nMean minus Median plot") fig3 = plt.figure() plt.title("Difference between median & mean in healthcare industry") plt.ylabel("Difference of median & mean value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), np.array(annual_mean) - np.array(annual_median)) plt.show()<jupyter_output>A total of 4998321 data pieces are related with healthcare industry The mean salary of the healthcare industry is 64665.53658618564 <jupyter_text>### 2. Analysis on companies<jupyter_code>if not company_analyzed: print("Companies in healthcare industries include: \n") print(sorted(pd.unique(healthcare_df['KF_ID']))) print("\nA total of " + str(len(sorted(pd.unique(healthcare_df['KF_ID'])))) + " companies") # Year distribution company_year_dict = defaultdict(list) year_company_dict = defaultdict(list) for company in sorted(pd.unique(healthcare_df['KF_ID'])): company_healthcare_df = healthcare_df[healthcare_df['KF_ID'] == company] for year in range(2008, 2020): if company_healthcare_df[company_healthcare_df['CalendarYear'] == year].shape[0] > 0: company_year_dict[int(company)].append(year) with open(os.path.join(root_dir, "task4", "Healthcare_Companies_Distribution_Across_Years.json"), 'w') as output_file: print("\nSaving the distribtion of companies across the years from 2008 to 2019 in json file") json.dump(company_year_dict, output_file) print("\nOf all healthcare companies, counting the number of years included in this survey\n") len_distribution = defaultdict(int) for company in company_year_dict: len_distribution[len(company_year_dict[company])] += 1 fig1 = plt.figure() plt.title("Healthcare companies surveyed, number of years") plt.xlabel("Years surveyed") plt.ylabel("Number of companies") plt.bar(list(len_distribution.keys()), list(len_distribution.values())) plt.show() print("\nDistribution of companies across the years") count_companies = defaultdict(int) for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] for company in pd.unique(year_healthcare_df['KF_ID']): count_companies[year] += 1 fig2 = plt.figure() plt.title("Healthcare companies surveyed across the years") plt.xlabel("Year") plt.ylabel("Number of companies") plt.bar(list(count_companies.keys()), list(count_companies.values())) plt.show() # Reference level distribution company_level_dict = defaultdict(list) for company in sorted(pd.unique(healthcare_df['KF_ID'])): company_healthcare_df = healthcare_df[healthcare_df['KF_ID'] == company] for level in pd.unique(company_healthcare_df['ReferenceLevelNum']): company_level_dict[int(company)].append(int(level)) with open(os.path.join(root_dir, "task4", "Healthcare_Companies_Distribution_Across_ReferLevel.json"), 'w') as output_file: print("\nSaving the distribtion of companies across reference levels in json file") json.dump(company_level_dict, output_file) print("\nOf all healthcare companies, counting the number of reference levels included in this survey\n") len_distribution = defaultdict(int) for company in company_level_dict: len_distribution[len(company_level_dict[company])] += 1 fig3 = plt.figure() plt.title("Healthcare companies surveyed, number of reference levels") plt.xlabel("Count on reference levels surveyed") plt.ylabel("Number of companies") plt.bar(list(len_distribution.keys()), list(len_distribution.values())) plt.show() # Incumbent point distribution company_point_dict = defaultdict(list) for company in sorted(pd.unique(healthcare_df['KF_ID'])): company_healthcare_df = healthcare_df[healthcare_df['KF_ID'] == company] for point in pd.unique(company_healthcare_df['IncumbentPointCount']): company_point_dict[int(company)].append(int(level)) with open(os.path.join(root_dir, "task4", "Healthcare_Companies_Distribution_Across_IncumbPoint.json"), 'w') as output_file: print("\nSaving the distribtion of companies across incumbent points in json file") json.dump(company_point_dict, output_file) print("\nOf all healthcare companies, counting the number of incumbent points included in this survey\n") len_distribution = defaultdict(int) for company in company_level_dict: len_distribution[len(company_point_dict[company])] += 1 fig3 = plt.figure() plt.title("Healthcare companies surveyed, number of incumbent points") plt.xlabel("Count on incmubent points surveyed") plt.ylabel("Number of companies") plt.bar(list(len_distribution.keys()), list(len_distribution.values())) plt.show() <jupyter_output>Companies in healthcare industries include: [19, 20, 22, 25, 26, 27, 30, 47, 48, 52, 53, 58, 60, 68, 69, 83, 93, 94, 97, 99, 100, 106, 109, 111, 113, 135, 139, 140, 145, 154, 155, 160, 162, 163, 165, 175, 188, 190, 191, 202, 203, 205, 207, 208, 210, 211, 212, 213, 214, 216, 217, 218, 221, 224, 228, 232, 236, 245, 246, 248, 249, 252, 254, 255, 256, 257, 258, 259, 260, 262, 263, 264, 275, 278, 290, 292, 296, 299, 301, 307, 311, 334, 344, 345, 347, 349, 351, 353, 355, 359, 368, 371, 372, 373, 381, 383, 385, 388, 391, 401, 403, 404, 420, 421, 422, 423, 424, 428, 429, 431, 439, 440, 449, 454, 455, 467, 483, 484, 486, 491, 493, 494, 506, 510, 519, 520, 521, 522, 531, 533, 538, 539, 544, 548, 559, 565, 569, 578, 579, 580, 581, 595, 624, 644, 650, 653, 657, 660, 674, 681, 687, 693, 709, 723, 724, 727, 742, 743, 746, 750, 755, 764, 773, 779, 780, 781, 784, 786, 787, 793, 814, 815, 819, 821, 825, 832, 833, 834, 836, 843, 857, 858, 859, 860, 862, 863, 864, 865, 866, 867, 868, 869, 870, 871, 872[...]<jupyter_text>### 3. Improved analysis by deleting companies that are only included in the survey for 1-2 years<jupyter_code># The following analysis gets rid of companies only taking 1 or 2 years of survey if not company_analyzed: new_healthcare_df = None for company in company_year_dict: if len(company_year_dict[company]) > min_surveying_years: new_healthcare_df = merge_dataframe(new_healthcare_df, healthcare_df[healthcare_df['KF_ID'] == company]) print("Of companies over " + str(min_surveying_years) + " years of survey\n") print("A total of " + str(new_healthcare_df.shape[0]) + " data pieces are related with healthcare industry\n") print("\nDistribution of companies across the years after improvement\n") count_companies = defaultdict(int) for year in range(2008, 2020): year_healthcare_df = new_healthcare_df[new_healthcare_df['CalendarYear'] == year] for company in pd.unique(year_healthcare_df['KF_ID']): count_companies[year] += 1 fig0 = plt.figure() plt.title("Healthcare companies surveyed across the years") plt.xlabel("Year") plt.ylabel("Number of companies") plt.bar(list(count_companies.keys()), list(count_companies.values())) plt.show() # Mean healthcare_mean_salary = new_healthcare_df['Base Salary'].mean() print("The mean salary of the healthcare industry is " + str(healthcare_mean_salary) + "\n") annual_mean = [] for year in range(2008, 2020): year_healthcare_df = new_healthcare_df[new_healthcare_df['CalendarYear'] == year] company_year_mean = year_healthcare_df[['KF_ID', 'Base Salary']].groupby(by = ['KF_ID']).mean() annual_mean_salary = company_year_mean['Base Salary'].mean() annual_mean.append(annual_mean_salary) fig1 = plt.figure() plt.title("The annual mean salary of healthcare industry") plt.ylabel("Mean salary value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), annual_mean) plt.show() # Median healthcare_median_salary = new_healthcare_df['Base Salary'].median() print("\nThe median salary of the healthcare industry is " + str(healthcare_median_salary) + "\n") annual_median = [] for year in range(2008, 2020): year_healthcare_df = new_healthcare_df[new_healthcare_df['CalendarYear'] == year] company_year_median = year_healthcare_df[['KF_ID', 'Base Salary']].groupby(by = ['KF_ID']).median() annual_median_salary = company_year_median['Base Salary'].median() annual_median.append(annual_median_salary) fig2 = plt.figure() plt.title("The annual median salary of healthcare industry") plt.ylabel("Median salary value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), annual_median) plt.show() print("\nMean minus Median plot") fig3 = plt.figure() plt.title("Difference between median & mean in healthcare industry") plt.ylabel("Difference of median & mean value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), np.array(annual_mean) - np.array(annual_median)) plt.show()<jupyter_output>Of companies over 3 years of survey A total of 4756131 data pieces are related with healthcare industry Distribution of companies across the years after improvement <jupyter_text>### Results comparison between raw & processed: The companies are deleted mainly in year 2012-2016 Not much difference found in year 2012-2016, but the mean & median value changes evidently in 2017-2019. ### 4. Improved analysis by deleting companies that are containing too few or too many reference levels<jupyter_code># The following analysis gets rid of companies having too few reference levels, or too many reference levels if not company_analyzed: new_healthcare_df = None for company in company_year_dict: if (len(company_year_dict[company]) > min_level_counts) and (len(company_year_dict[company]) < max_level_counts): new_healthcare_df = merge_dataframe(new_healthcare_df, healthcare_df[healthcare_df['KF_ID'] == company]) print("Of companies over " + str(min_surveying_years) + " years of survey\n") print("A total of " + str(new_healthcare_df.shape[0]) + " data pieces are related with healthcare industry\n") # Mean healthcare_mean_salary = new_healthcare_df['Base Salary'].mean() print("The mean salary of the healthcare industry is " + str(healthcare_mean_salary) + "\n") annual_mean = [] for year in range(2008, 2020): year_healthcare_df = new_healthcare_df[new_healthcare_df['CalendarYear'] == year] company_year_mean = year_healthcare_df[['KF_ID', 'Base Salary']].groupby(by = ['KF_ID']).mean() annual_mean_salary = company_year_mean['Base Salary'].mean() annual_mean.append(annual_mean_salary) fig1 = plt.figure() plt.title("The annual mean salary of healthcare industry") plt.ylabel("Mean salary value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), annual_mean) plt.show() # Median healthcare_median_salary = new_healthcare_df['Base Salary'].median() print("\nThe median salary of the healthcare industry is " + str(healthcare_median_salary) + "\n") annual_median = [] for year in range(2008, 2020): year_healthcare_df = new_healthcare_df[new_healthcare_df['CalendarYear'] == year] company_year_median = year_healthcare_df[['KF_ID', 'Base Salary']].groupby(by = ['KF_ID']).median() annual_median_salary = company_year_median['Base Salary'].median() annual_median.append(annual_median_salary) fig2 = plt.figure() plt.title("The annual median salary of healthcare industry") plt.ylabel("Median salary value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), annual_median) plt.show() print("\nMean minus Median plot") fig3 = plt.figure() plt.title("Difference between median & mean in healthcare industry") plt.ylabel("Difference of median & mean value in dollar") plt.xlabel("Year") plt.plot(range(2008, 2020), np.array(annual_mean) - np.array(annual_median)) plt.show()<jupyter_output>Of companies over 3 years of survey A total of 4439477 data pieces are related with healthcare industry The mean salary of the healthcare industry is 65123.505266949236 <jupyter_text>### Results comparison between raw & processed: Not much difference found in year 2012-2016, but the mean & median value changes evidently in 2017-2019. We can conclude that many conpanies only participating into the survey in year 2017-2019, influencing the quality of data. Similar to changes from the improvements of deleting companies of 1-2 years of enrollment.### 5. Relevance analysis among variables<jupyter_code># Only counting ReferenceLevelNum and IncumbentPointCount into regression var_df = healthcare_df[['ReferenceLevelNum', 'IncumbentPointCount']] salary_df = healthcare_df['Base Salary'] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) linear_reg = LinearRegression() linear_reg.fit(train_var, train_salary) pred_salary = linear_reg.predict(test_var) print("The coeficients of linear regression are " + str(linear_reg.coef_)) print("The intercept of linear regresssion is " + str(linear_reg.intercept_) + "\n") print("The R2 score: ") print("%.3f\n" % (r2_score(test_salary, pred_salary))) print("The RMSE error of regression model is %.3f" % (np.sqrt(mean_squared_error(test_salary, pred_salary)))) # Counting ReferenceLevelNum, IncumbentPointCount, Num into regression var_df = healthcare_df[['ReferenceLevelNum', 'IncumbentPointCount', 'NumOfEmpDesc', 'Base Salary']].dropna() salary_df = var_df['Base Salary'] var_df = var_df[['ReferenceLevelNum', 'IncumbentPointCount', 'NumOfEmpDesc']] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) linear_reg = LinearRegression() linear_reg.fit(train_var, train_salary) pred_salary = linear_reg.predict(test_var) print("The coeficients of linear regression are " + str(linear_reg.coef_)) print("The intercept of linear regresssion is " + str(linear_reg.intercept_) + "\n") print("The R2 score: ") print("%.3f\n" % (r2_score(test_salary, pred_salary))) print("The RMSE error of regression model is %.3f" % (np.sqrt(mean_squared_error(test_salary, pred_salary))))<jupyter_output>The coeficients of linear regression are [ 641.50014431 159.16886281 -427.24991186] The intercept of linear regresssion is 6193.7344144186645 The R2 score: 0.569 The RMSE error of regression model is 22538.013 <jupyter_text>### Results comparison befroe & after NumOfEmpDesc added to regression: The R2 score and RMSE error do not improve very much after the factor of Number of employees added. We can conclude that NumOfEmpDesc is not primary factor affecting Base Salary.## Reference Level Analysis ### 1. Raw analysis of all levels in healthcare sector<jupyter_code>if not reference_level_analyzed: level_count = defaultdict(int) level_mean = defaultdict(float) level_median = defaultdict(float) for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): level_healthcare_df = healthcare_df[healthcare_df['ReferenceLevelNum'] == level] level_count[int(level)] = np.log10(level_healthcare_df.shape[0]) company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() level_mean[int(level)] = company_level_mean.mean() company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() level_median[int(level)] = company_level_median.median() print("\nDistribution across reference levels\n") fig1 = plt.figure() plt.title("Reference level distribution of healthcare industry") plt.ylabel("Log scaled number of data points") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_count.values())) plt.show() print("\nMean value of salary across reference levels\n") fig2 = plt.figure() plt.title("Mean Salary across reference level of healthcare industry") plt.ylabel("Mean salary of data points") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_mean.values())) plt.show() print("\nMedian values across reference levels\n") fig3 = plt.figure() plt.title("Median Salary across reference level of healthcare industry") plt.ylabel("Median salary of data points") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_median.values())) plt.show() <jupyter_output> Distribution across reference levels <jupyter_text>### Results from reference level analysis: The data points with level greater than 33 is not accurate. Level 99 is not applicable, so delete this level in all analysis in this notebook.### 2. Yearly analysis on reference level of healthcare industry<jupyter_code>if not reference_level_analyzed: print("\nDistribution across reference levels\n") fig1 = plt.figure(figsize=(8,4)) plt.title("Reference level distribution of healthcare industry") plt.ylabel("Percentage of data points") plt.xlabel("Reference Level") for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] year_level_count = defaultdict(lambda: defaultdict(int)) for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): level_healthcare_df = year_healthcare_df[year_healthcare_df['ReferenceLevelNum'] == level] year_level_count[int(level)] = level_healthcare_df.shape[0] all_counts = sum(year_level_count.values()) for level in year_level_count: year_level_count[level] /= all_counts plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_count.values()), label = str(year)) plt.legend() plt.show() print("\nMean value of salary across reference levels\n") fig2 = plt.figure(figsize=(8,4)) plt.title("Mean Salary across reference level of healthcare industry") plt.ylabel("Mean salary of data points") plt.xlabel("Reference Level") for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] year_level_mean = defaultdict(lambda: defaultdict(float)) for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): level_healthcare_df = year_healthcare_df[year_healthcare_df['ReferenceLevelNum'] == level] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() year_level_mean[int(level)] = company_level_mean.mean() plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_mean.values()), label = str(year)) plt.legend() plt.show() print("\nMedian values across reference levels\n") fig3 = plt.figure(figsize=(8,4)) plt.title("Median Salary across reference level of healthcare industry") plt.ylabel("Median salary of data points") plt.xlabel("Reference Level") for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] year_level_median = defaultdict(lambda: defaultdict(float)) for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): level_healthcare_df = year_healthcare_df[year_healthcare_df['ReferenceLevelNum'] == level] company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() year_level_median[int(level)] = company_level_median.median() plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_median.values()), label = str(year)) plt.legend() plt.show() <jupyter_output> Distribution across reference levels <jupyter_text>### Results from yearly reference level analysis: From the year 2017, the mean & median values of level 10-25 falls greatly, which explains the downgoing values in the previous plot of company-year relationship Data in year 2012 might contain some errors, since the mean & median salary of high reference level is comparably lower than those in other years.### 3. Regression analysis on reference level of healthcare industry<jupyter_code>if not reference_level_analyzed: # Linear regression on reference level and salary var_df = healthcare_df[['ReferenceLevelNum']] salary_df = healthcare_df['Base Salary'] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) linear_reg = LinearRegression() linear_reg.fit(train_var, train_salary) pred_salary = linear_reg.predict(test_var) print("The coeficients of linear regression are " + str(linear_reg.coef_)) print("The intercept of linear regresssion is " + str(linear_reg.intercept_) + "\n") print("The R2 score: ") print("%.3f\n" % (r2_score(test_salary, pred_salary))) print("The RMSE error of regression model is %.3f" % (np.sqrt(mean_squared_error(test_salary, pred_salary)))) if not reference_level_analyzed: # Polynomial regression on reference level r2_scores = [] RMSEs = [] min_deg = 1 max_deg = 10 for degree in range(min_deg, max_deg+1): var_df = healthcare_df[['ReferenceLevelNum']] salary_df = healthcare_df['Base Salary'] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) linear_reg = LinearRegression() train_var = PolynomialFeatures(degree).fit_transform(train_var) test_var = PolynomialFeatures(degree).fit_transform(test_var) linear_reg.fit(train_var, train_salary) pred_salary = linear_reg.predict(test_var) r2_scores.append(r2_score(test_salary, pred_salary)) RMSEs.append(np.sqrt(mean_squared_error(test_salary, pred_salary))) fig1 = plt.figure() plt.title("Regression analysis across reference level of healthcare industry") plt.ylabel("R2 score") plt.xlabel("Degree of regression") plt.xticks(range(min_deg, max_deg+1, 1)) plt.plot(range(min_deg, max_deg+1), r2_scores) plt.show() fig2 = plt.figure() plt.title("Regression analysis across reference level of healthcare industry") plt.ylabel("RMSE") plt.xlabel("Degree of regression") plt.xticks(range(min_deg, max_deg+1, 1)) plt.plot(range(min_deg, max_deg+1), RMSEs) plt.show() if not reference_level_analyzed: level_mean = defaultdict(float) level_median = defaultdict(float) for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): level_healthcare_df = healthcare_df[healthcare_df['ReferenceLevelNum'] == level] level_mean[int(level)] = level_healthcare_df['Base Salary'].dropna().mean() level_median[int(level)] = level_healthcare_df['Base Salary'].dropna().median() var_df = healthcare_df[['ReferenceLevelNum']] salary_df = healthcare_df['Base Salary'] linear_reg = LinearRegression() var_df = PolynomialFeatures(6).fit_transform(var_df) linear_reg.fit(var_df, salary_df) pred_result = [] for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): pred = [[level]] pred = PolynomialFeatures(6).fit_transform(pred) pred_result.append(linear_reg.predict(pred)) fig1 = plt.figure() plt.title("Approximation result of regression") plt.ylabel("Salary of data points") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_mean.values()), label = "Mean") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_median.values()), label = "Median") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), pred_result, label = "Regression") plt.legend() plt.show()<jupyter_output><empty_output><jupyter_text>### Results from regression analysis: The linear regression model is not complex enough to generalize the salary trend based on reference level. Considering the balance between model complexity and running time, the polynomial model with degree 5 or 6 could best describe the trend between reference level and base salary. The regression could predict the salary of low & medium reference level very well, but for high level incomes, the model is biased in prediction since the datapoints of high incomer is too few to make good predictions.## Incumbent Point Analysis ### 1. Raw analysis of all levels in healthcare sector <jupyter_code>if not incumbent_point_analyzed: level_count = defaultdict(int) level_mean = defaultdict(float) level_median = defaultdict(float) for level in sorted(pd.unique(healthcare_df['IncumbentPointCount'])): level_healthcare_df = healthcare_df[healthcare_df['IncumbentPointCount'] == level] level_count[int(level)] = level_healthcare_df.shape[0] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() level_mean[int(level)] = company_level_mean.mean() company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() level_median[int(level)] = company_level_median.median() all_counts = sum(level_count.values()) for level in level_count: level_count[level] /= all_counts print("\nDistribution across incumbent points\n") fig1 = plt.figure(figsize=(20,5)) plt.title("Incumbent point distribution of healthcare industry") plt.ylabel("Percentage of data points") plt.xlabel("Incumbent Point") plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(level_count.values())) plt.show() print("\nMean value of salary across incumbent points\n") fig2 = plt.figure(figsize=(20,5)) plt.title("Mean Salary across incumbent points of healthcare industry") plt.ylabel("Mean salary of data points") plt.xlabel("Incumbent Point") plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(level_mean.values())) plt.show() print("\nMedian values across incumbent points\n") fig3 = plt.figure(figsize=(20,5)) plt.title("Median Salary across incumbent points of healthcare industry") plt.ylabel("Median salary of data points") plt.xlabel("Incumbent Point") plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(level_median.values())) plt.show()<jupyter_output> Distribution across incumbent points <jupyter_text>### Results from incumbent points analysis: Similar to the conclusion in reference level, the incumbent point greater than 8000 is not accurate### 2. Yearly analysis on incumbent points of healthcare industry<jupyter_code>if not incumbent_point_analyzed: print("\nDistribution across incumbent points\n") fig1 = plt.figure(figsize=(50,5)) plt.title("Incumbent point distribution of healthcare industry") plt.ylabel("Percentage of data points") plt.xlabel("Incumbent Point") for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] year_level_count = defaultdict(lambda: defaultdict(int)) for level in sorted(pd.unique(healthcare_df['IncumbentPointCount'])): level_healthcare_df = year_healthcare_df[year_healthcare_df['IncumbentPointCount'] == level] year_level_count[int(level)] = level_healthcare_df.shape[0] all_counts = sum(level_count.values()) for level in year_level_count: year_level_count[level] /= all_counts plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(year_level_count.values()), label = str(year)) plt.legend() plt.show() print("\nMean value of salary across incumbent points\n") fig2 = plt.figure(figsize=(50,5)) plt.title("Mean Salary across incumbent point of healthcare industry") plt.ylabel("Mean salary of data points") plt.xlabel("Incumbent Point") for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] year_level_mean = defaultdict(lambda: defaultdict(float)) for level in sorted(pd.unique(healthcare_df['IncumbentPointCount'])): level_healthcare_df = year_healthcare_df[year_healthcare_df['IncumbentPointCount'] == level] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() year_level_mean[int(level)] = company_level_mean.mean() plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(year_level_mean.values()), label = str(year)) plt.legend() plt.show() print("\nMedian values across incumbent points\n") fig3 = plt.figure(figsize=(50,5)) plt.title("Median Salary across incumbent point of healthcare industry") plt.ylabel("Median salary of data points") plt.xlabel("Incumbent Point") for year in range(2008, 2020): year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == year] year_level_median = defaultdict(lambda: defaultdict(float)) for level in sorted(pd.unique(healthcare_df['IncumbentPointCount'])): level_healthcare_df = year_healthcare_df[year_healthcare_df['IncumbentPointCount'] == level] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() year_level_median[int(level)] = company_level_median.median() plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(year_level_median.values()), label = str(year)) plt.legend() plt.show()<jupyter_output> Distribution across incumbent points <jupyter_text>### Results from yearly incumbent points analysis: The incubent point is too detailed, causing too few data points covered in one incumbent value. This is not suitable for yearly analysis.### 3. Regression analysis on incumbent points of healthcare industry<jupyter_code>if not incumbent_point_analyzed: # Linear regression on reference level and salary var_df = healthcare_df[['IncumbentPointCount']] salary_df = healthcare_df['Base Salary'] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) linear_reg = LinearRegression() linear_reg.fit(train_var, train_salary) pred_salary = linear_reg.predict(test_var) print("The coeficients of linear regression are " + str(linear_reg.coef_)) print("The intercept of linear regresssion is " + str(linear_reg.intercept_) + "\n") print("The R2 score: ") print("%.3f\n" % (r2_score(test_salary, pred_salary))) print("The RMSE error of regression model is %.3f" % (np.sqrt(mean_squared_error(test_salary, pred_salary)))) if not incumbent_point_analyzed: # Polynomial regression on reference level r2_scores = [] RMSEs = [] min_deg = 1 max_deg = 10 for degree in range(min_deg, max_deg+1): var_df = healthcare_df[['IncumbentPointCount']] salary_df = healthcare_df['Base Salary'] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) linear_reg = LinearRegression() train_var = PolynomialFeatures(degree).fit_transform(train_var) test_var = PolynomialFeatures(degree).fit_transform(test_var) linear_reg.fit(train_var, train_salary) pred_salary = linear_reg.predict(test_var) r2_scores.append(r2_score(test_salary, pred_salary)) RMSEs.append(np.sqrt(mean_squared_error(test_salary, pred_salary))) fig1 = plt.figure() plt.title("Regression analysis across incumbent points of healthcare industry") plt.ylabel("R2 score") plt.xlabel("Degree of regression") plt.xticks(range(min_deg, max_deg+1, 1)) plt.plot(range(min_deg, max_deg+1), r2_scores) plt.show() fig2 = plt.figure() plt.title("Regression analysis across incumbent points of healthcare industry") plt.ylabel("RMSE") plt.xlabel("Degree of regression") plt.xticks(range(min_deg, max_deg+1, 1)) plt.plot(range(min_deg, max_deg+1), RMSEs) plt.show() if not incumbent_point_analyzed: level_mean = defaultdict(float) level_median = defaultdict(float) for level in sorted(pd.unique(healthcare_df['IncumbentPointCount'])): level_healthcare_df = healthcare_df[healthcare_df['IncumbentPointCount'] == level] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() level_mean[int(level)] = company_level_mean.mean() company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() level_median[int(level)] = company_level_median.median() var_df = healthcare_df[['IncumbentPointCount']] salary_df = healthcare_df['Base Salary'] linear_reg = LinearRegression() var_df = PolynomialFeatures(5).fit_transform(var_df) linear_reg.fit(var_df, salary_df) pred_result = [] for level in sorted(pd.unique(healthcare_df['IncumbentPointCount'])): pred = [[level]] pred = PolynomialFeatures(5).fit_transform(pred) pred_result.append(linear_reg.predict(pred)) fig1 = plt.figure() plt.title("Approximation result of regression") plt.ylabel("Salary of data points") plt.xlabel("Incumbent Points") plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(level_mean.values()), label = "Mean") plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), list(level_median.values()), label = "Median") plt.plot(sorted(pd.unique(healthcare_df['IncumbentPointCount'])), pred_result, label = "Regression") plt.legend() plt.show()<jupyter_output><empty_output><jupyter_text>### Results from regression analysis: The linear regression model is not complex enough to generalize the salary trend based on incumbent points. Considering the balance between model complexity and running time, the polynomial model with degree 5 could best describe the trend between reference level and base salary. If the model degree is higher than 6, then the model will produce a much more unreliable result due to high model complexity & uneven distribution of datapoints. The regression could predict the salary of low & medium incumbent points very well. Similar to the problems in reference level, the model fails to predict high incumbent level incomes since datapoints of those high level incomers are too rare.## Total Regression Analysis ### 1. Multinomial Regression of all variables<jupyter_code>var_df = healthcare_df[['ReferenceLevelNum', 'IncumbentPointCount']] salary_df = healthcare_df['Base Salary'] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) linear_reg = LinearRegression() train_var = PolynomialFeatures(5).fit_transform(train_var) test_var = PolynomialFeatures(5).fit_transform(test_var) linear_reg.fit(train_var, train_salary) pred_salary = linear_reg.predict(test_var) r2_scores.append(r2_score(test_salary, pred_salary)) RMSEs.append(np.sqrt(mean_squared_error(test_salary, pred_salary))) print("The coeficients of linear regression are " + str(linear_reg.coef_)) print("The intercept of linear regresssion is " + str(linear_reg.intercept_) + "\n") print("The R2 score: ") print("%.3f\n" % (r2_score(test_salary, pred_salary))) print("The RMSE error of regression model is %.3f" % (np.sqrt(mean_squared_error(test_salary, pred_salary))))<jupyter_output>The coeficients of linear regression are [ 0.00000000e+00 -4.58595334e-01 -5.97359644e-01 -6.49593640e-01 -5.57689876e+00 -1.13404422e+00 -1.88283369e+00 -2.09317582e+01 8.12706609e-01 -3.92779306e-03 2.03037016e+01 9.13369895e-01 -4.90123411e-02 2.81888286e-04 -9.20061909e-08 -8.35500798e-01 -1.58848082e-03 7.13899677e-04 -4.67727359e-06 1.83201986e-09 1.88225722e-12] The intercept of linear regresssion is 41279.602532898396 The R2 score: 0.589 The RMSE error of regression model is 21922.198 <jupyter_text>### Results from polynomial regression analysis: The multinomial regression model does not effectively improve the performance compared to linear model.### 2. Regresssion with Supportive Vector of all variables<jupyter_code>sv_regression = SVR(kernel='linear', degree = 1, max_iter = 10000) var_df = healthcare_df[['ReferenceLevelNum', 'IncumbentPointCount']] salary_df = healthcare_df['Base Salary'] train_var, test_var, train_salary, test_salary = train_test_split(var_df, salary_df, test_size = 0.2) sv_regression.fit(train_var, train_salary) pred_salary = sv_regression.predict(test_var) print("The coeficients of linear regression are " + str(linear_reg.coef_)) print("The intercept of linear regresssion is " + str(linear_reg.intercept_) + "\n") print("The R2 score: ") print("%.3f\n" % (r2_score(test_salary, pred_salary))) print("The RMSE error of regression model is %.3f" % (np.sqrt(mean_squared_error(test_salary, pred_salary))))<jupyter_output><empty_output><jupyter_text>### Results from Supportive Vector regression analysis: The SV-regression is very slow in running, and it often fails to converge to a small tolerance limit. And the model performance is worse than traditional regression. Maybe this kind of regression could have better performance, but it may need searching on better hyper-parameters for good results on regression.## Skill level Analysis<jupyter_code>if not skill_level_analyzed: print("\nComparison of reference levels of total and 2017\n") level_count = defaultdict(int) level_mean = defaultdict(float) level_median = defaultdict(float) for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): level_healthcare_df = healthcare_df[healthcare_df['ReferenceLevelNum'] == level] level_count[int(level)] = level_healthcare_df.shape[0] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() level_mean[int(level)] = company_level_mean.mean() company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() level_median[int(level)] = company_level_median.median() counts = sum(level_count.values()) for level in level_count: level_count[level] /= counts year_healthcare_df = healthcare_df[healthcare_df['CalendarYear'] == 2017] year_level_count = defaultdict(lambda: defaultdict(int)) year_level_mean = defaultdict(float) year_level_median = defaultdict(float) for level in sorted(pd.unique(healthcare_df['ReferenceLevelNum'])): level_healthcare_df = year_healthcare_df[year_healthcare_df['ReferenceLevelNum'] == level] year_level_count[int(level)] = level_healthcare_df.shape[0] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() year_level_mean[int(level)] = company_level_mean.mean() company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() year_level_median[int(level)] = company_level_median.median() counts = sum(year_level_count.values()) for level in year_level_count: year_level_count[level] /= counts fig1 = plt.figure() plt.title("Reference level distribution of healthcare industry") plt.ylabel("Percentage of total number") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_count.values()), label = 'mean') plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_count.values()), label = '2017') plt.legend() plt.show() fig2 = plt.figure() plt.title("Income distribution across reference level of healthcare industry") plt.ylabel("Mean income") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_mean.values()), label = 'mean') plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_mean.values()), label = '2017') plt.legend() plt.show() fig3 = plt.figure() plt.title("Income distribution across reference level of healthcare industry") plt.ylabel("Median income") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(level_median.values()), label = 'median') plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_median.values()), label = '2017') plt.legend() plt.show() <jupyter_output> Comparison of reference levels of total and 2017 <jupyter_text>### Results from distribution of total and 2017: The distribution of all incomers across different levels in 2017 are basically the same as the general one. Medium & high skill level salaries has risen in the year of 2017. This should be caused by events of mid & high skills technicans in year 2017. <jupyter_code>if not skill_level_analyzed: year_companies = pd.unique(year_healthcare_df['KF_ID']) company_level_median = defaultdict(int) companies_df = None for company in year_companies: company_data_df = healthcare_df[healthcare_df['KF_ID'] == company] companies_df = merge_dataframe(companies_df, company_data_df) company_2017_level_count = defaultdict(int) company_2017_level_mean = defaultdict(float) company_2017_level_median = defaultdict(float) for level in sorted(pd.unique(companies_df['ReferenceLevelNum'])): level_healthcare_df = companies_df[companies_df['ReferenceLevelNum'] == level] company_2017_level_count[int(level)] = level_healthcare_df.shape[0] company_level_mean = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).mean() company_2017_level_mean[int(level)] = company_level_mean.mean() company_level_median = level_healthcare_df[['KF_ID', 'Base Salary']].dropna().groupby(by = ['KF_ID']).median() company_2017_level_median[int(level)] = company_level_median.median() counts = sum(company_2017_level_count.values()) for level in company_2017_level_count: company_2017_level_count[level] /= counts fig1 = plt.figure() plt.title("Reference level distribution of healthcare companies surveyed in 2017") plt.ylabel("Percentage of total number") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(company_2017_level_count.values()), label = 'all count') plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_count.values()), label = '2017') plt.legend() plt.show() fig2 = plt.figure() plt.title("Income distribution across reference level of healthcare companies surveyed in 2017") plt.ylabel("Mean income") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(company_2017_level_mean.values()), label = 'all mean') plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_mean.values()), label = '2017') plt.legend() plt.show() fig3 = plt.figure() plt.title("Income distribution across reference level of healthcare companies surveyed in 2017") plt.ylabel("Median income") plt.xlabel("Reference Level") plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(company_2017_level_median.values()), label = 'all median') plt.plot(sorted(pd.unique(healthcare_df['ReferenceLevelNum'])), list(year_level_median.values()), label = '2017') plt.legend() plt.show()<jupyter_output><empty_output>
no_license
/healthcare_sector_analysis.ipynb
rsk2327/PDSG_PayInequality
16
<jupyter_start><jupyter_text><jupyter_code>from google.colab import drive drive.mount('/content/drive') from google.colab.patches import cv2_imshow import cv2 import numpy as np import os import math #drive/MyDrive/new/yolo/ # Load Yolo #drive/MyDrive/new/yolo/ net = cv2.dnn.readNet("drive/MyDrive/new/yolo/yolov3.weights", "drive/MyDrive/new/yolo/yolov3.cfg") classes = [] with open("drive/MyDrive/new/yolo/coco.names", "r") as f: classes = [line.strip() for line in f.readlines()] layer_names = net.getLayerNames() output_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()] colors = np.random.uniform(0, 255, size=(len(classes), 3)) # myset=set() # for id in class_ids: # myset.add(id) # myset #firstFrame firstFrame =None first=1 first2 =1 warn =0; #variable to indicate warning #vediocapture vediocap = cv2.VideoCapture('drive/MyDrive/new/yolo/chair2.mp4') vediocap.set(cv2.CAP_PROP_FRAME_WIDTH,640) vediocap.set(cv2.CAP_PROP_FRAME_HEIGHT,480) vediocap.set(cv2.CAP_PROP_FPS, 30) # Define the codec and create VideoWriter object fourcc = cv2.VideoWriter_fourcc(*'DIVX') outer = cv2.VideoWriter('char2.avi',fourcc, 30.0, (640,480)) terminate =0; fmid_x=[] fmid_y=[] mid_x=[] mid_y=[] def distance(x1,x2,y1,y2): distc = math.sqrt((x2 - x1)**2 + (y2 - y1)**2) return distc while vediocap.isOpened(): # Capture frame-by-frame ret, img = vediocap.read() if not ret: break # Loading image # img = cv2.imread("drive/MyDrive/new/yolo/") # img = cv2.resize(img, None, fx=0.4, fy=0.4) height,width,channels = img.shape #Detecting objects blob = cv2.dnn.blobFromImage(img, 0.00392, (416, 416), (0, 0, 0), True, crop=False) net.setInput(blob) outs = net.forward(output_layers) # Showing informations on the screen class_ids = [] confidences = [] boxes = [] myset =set() for out in outs: for detection in out: scores = detection[5:] class_id = np.argmax(scores) confidence = scores[class_id] if confidence > 0.5: # Object detected center_x = int(detection[0] * width) center_y = int(detection[1] * height) w = int(detection[2] * width) h = int(detection[3] * height) # Rectangle coordinates x = int(center_x - w / 2) y = int(center_y - h / 2) boxes.append([x, y, w, h]) confidences.append(float(confidence)) class_ids.append(class_id) #here we add set myset.add(class_id) #if object is Lost #if live we update for specific time if first: print("first frame") fixedset=set() first =0 firstframe =img fixedset =[c for c in myset] no_of_fixedset =len(fixedset) # print("fixedSet ",len(fixedset)) # print("myset ",len(myset)) no_of_myset = len(myset) if(no_of_fixedset != no_of_myset): warn=warn+1; entered =0 lost=0 if(warn >5): warn=0 #update the values print("=======================================================") if(no_of_fixedset < no_of_myset): #if both no. equal it will not come here print("New object entered ") entered=1 else: print("object Lost") lost=1 no_of_fixedset = no_of_myset indexes = cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.4) kil =len(boxes) for j in range(kil): fmid_x.append(0) fmid_y.append(0) mid_x.append(0) mid_y.append(0) font = cv2.FONT_HERSHEY_PLAIN for i in range(len(boxes)): # print(i) k=0 if i in indexes: x, y, w, h = boxes[i] label = str(classes[class_ids[i]]) if(entered): label =label+" Entered " entered =0 if(lost): label =label+" MISSING ! ALERT " lost =0 color = colors[i] cv2.rectangle(img, (x, y), (x + w, y + h), color, 2) cv2.putText(img, label, (x, y + 30), font, 3, color, 3) #if object moves if kil: fmid_x[k] =int((2*x+w)/2) fmid_y[k] =int((2*y+h)/2) kil =kil-1 mid_x[k] =(2*x+w)/2 mid_y[k] =(2*y+h)/2 #find distance b/w two objects dist = distance(fmid_x[k],mid_x[k],fmid_y[k],mid_y[k]) #using this we can even predict direct of object moving print("distance : " + label,dist) k = k+1 # cv2_imshow(img) #uncoment this to see the image frames outer.write(img) terminate=terminate+1 vediocap.release() outer.release() cv2.waitKey(0) # cv2.destroyAllWindows() <jupyter_output>first frame distance : chair 0.7071067811865476 distance : book 0.7071067811865476 distance : chair 0.7071067811865476 distance : book 0.7071067811865476 distance : chair 0.5 distance : book 0.5 distance : chair 0.5 distance : book 0.5 distance : chair 0.7071067811865476 distance : book 0.5 distance : chair 0.7071067811865476 distance : book 0.0 distance : chair 0.7071067811865476 distance : book 0.5 distance : chair 0.0 distance : book 0.5 distance : chair 0.7071067811865476 distance : book 0.5 distance : chair 0.5 distance : book 0.5 distance : chair 0.0 distance : book 0.7071067811865476 distance : chair 0.7071067811865476 distance : book 0.5 distance : person 0.7071067811865476 distance : chair 0.7071067811865476 distance : book 0.0 distance : person 0.0 distance : chair 0.5 distance : book 0.0 distance : person 0.5 distance : chair 0.5 distance : book 0.0 distance : person 0.0 distance : chair 0.5 distance : person 0.0 distance : book 0.0 ==========================================[...]
no_license
/YoloVedioCap.ipynb
brkuhgk/yolo
1
<jupyter_start><jupyter_text># Assignment 01Made by : Neeraj kumar Shift : 03:30 P.M TO 06:30 P.M<jupyter_code>#\newline Backslash and newline ignored print("Line1 \n Line2 \n Line3") #Example print("\\") #Backslash(\\) is used to print single backslash and if we want to print four backslash so we must used eight #backslash at a single time. In short double backslash is equals to single backslash(\\ = \) print('\'') #if we want to print single qoute so we must write like this. Another Example. print('I \'m Neeraj') print("\"") #For double qoute, we must write like this.<jupyter_output>" <jupyter_text>If we want to use double with single qoute so we can. for example : print("My name is 'neeraj'") but we can't used double qoute with same double qoute, then it will be give error for example : print("My name is "NAME"") <jupyter_code>print("My name is 'Name'") #So like this we can't used double qoute with as same doube qoute but we can use single qoute with double qoute print("My name is "name") print("Hello \"World\" World") #If we want to use double qoute with double qoute as well, so we must used backslash like this. and this is the Escape Sequence print("Hello \t World")# \t Backslash "t" is used for horizantal tab space between charactors. print("Hell\blo") #(\b) backslash b is used to finish the extra words. like this. #Date: 19.05.2019 a = a + 5 a = 10 a + 5 a a = a +5 a a += 5 a a -= 5 a a *= 5 a a/= 5 a<jupyter_output><empty_output><jupyter_text># Comparison Operator<jupyter_code>a = 2 b = 2 a == b a!= b #not eqaul a > b a< b a >= b a <= b a b a <= b a = 2 b = 2 a <= b<jupyter_output><empty_output><jupyter_text># Math Expression: Eliminatting ambiuity<jupyter_code>10 ** 2 10 * 3 ** 2 + (12 * 3)/ 2 2 / 4 - 3 * 9 + 2<jupyter_output><empty_output>
no_license
/Assignment 01 Sunday 03 30 to 06 30.ipynb
neeraj72/new
4
<jupyter_start><jupyter_text>## ReLU<jupyter_code>model_metadata_file_path = pathlib.Path("..", training_logs["relu"], "metadata.json") model_metadata = io.read_metadata(str(model_metadata_file_path)) model_metadata history = model_metadata["history"] metrics_df = visualize.create_metrics_dataframe(history) loss_plot = visualize.learning_curves("loss", metrics_df) visualize.save_plot(loss_plot, f"../reports/figures/{species}-relu.loss.png") accuracy_plot = visualize.learning_curves("accuracy", metrics_df) visualize.save_plot(accuracy_plot, f"../reports/figures/{species}-relu.accuracy.png")<jupyter_output><empty_output><jupyter_text>## ReLU with Dropout<jupyter_code>dropout_model_metadata_file_path = pathlib.Path("..", training_logs["relu-dropout"], "metadata.json") dropout_model_metadata = io.read_metadata(str(dropout_model_metadata_file_path)) dropout_model_metadata dropout_history = dropout_model_metadata["history"] dropout_metrics_df = visualize.create_metrics_dataframe(dropout_history) dropout_loss_plot = visualize.learning_curves("loss", dropout_metrics_df) visualize.save_plot(dropout_loss_plot, f"../reports/figures/{species}-relu-dropout.loss.png") dropout_accuracy_plot = visualize.learning_curves("accuracy", dropout_metrics_df) visualize.save_plot(dropout_accuracy_plot, f"../reports/figures/{species}-relu.accuracy.png")<jupyter_output><empty_output>
permissive
/notebooks/4.1-agj-evaluate-Tomato.ipynb
bitjockey42/ikapati-research
2
<jupyter_start><jupyter_text># Gradient Descent: Step Sizes - Lab## IntroductionIn this lab, we'll practice applying gradient descent. As we know gradient descent begins with an initial regression line, and moves to a "best fit" regression line by changing values of $m$ and $b$ and evaluating the RSS. So far, we have illustrated this technique by changing the values of $m$ and evaluating the RSS. In this lab, we will work through applying our technique by changing the value of $b$ instead. Let's get started.## ObjectivesYou will be able to: - Understand how to go from RSS to finding a "best fit" line - Understand how gradient descent can be used to find the best intercept for your linear regression model## Setting up our initial regression lineOnce again, we'll take take a look at revenues our data example, which looks like this:<jupyter_code>import numpy as np np.set_printoptions(formatter={'float_kind':'{:f}'.format}) import matplotlib.pyplot as plt np.random.seed(225) x = np.random.rand(30, 1).reshape(30) y_randterm = np.random.normal(0,3,30) y = 3+ 50* x + y_randterm plt.plot(x, y, '.b') plt.xlabel("x", fontsize=14) plt.ylabel("y", fontsize=14); plt.show()<jupyter_output><empty_output><jupyter_text>We can start with some values for an initial not-so-accurate regression line, $y = 43x + 12$.<jupyter_code>def regression_formula(x): return 12 + 43*x np.random.seed(225) x = np.random.rand(30,1).reshape(30) y_randterm = np.random.normal(0,3,30) y = 3+ 50* x + y_randterm plt.plot(x, y, '.b') plt.plot(x, regression_formula(x), '-') plt.xlabel("x", fontsize=14) plt.ylabel("y", fontsize=14); def errors(x_values, y_values, m, b): y_line = (b + m*x_values) return (y_values - y_line) def squared_errors(x_values, y_values, m, b): return errors(x_values, y_values, m, b)**2 def residual_sum_squares(x_values, y_values, m, b): return sum(squared_errors(x_values, y_values, m, b))<jupyter_output><empty_output><jupyter_text>Now using the `residual_sum_squares`, function, we calculate the RSS to measure the accuracy of the regression line to our data. Let's take another look at that function:<jupyter_code>residual_sum_squares(x, y , 43, 12) <jupyter_output><empty_output><jupyter_text>### Building a cost curveNow let's use the `residual_sum_squares` function to build a cost curve. Keeping the $m$ value fixed at $43$, write a function called `rss_values`. * `rss_values` passes our dataset with the `x_values` and `y_values` arguments. * It also takes a list of values of $b$, and an initial $m$ value as arguments. * It outputs a numpy array with a first column of `b_values` and `rss_values`, with each key pointing to a list of the corresponding values.<jupyter_code>def rss_values(x_values, y_values, m, b_values): B_RSS = [] for bval in b_values: rss = residual_sum_squares(x_values, y_values, m, bval) B_RSS.append([bval, rss]) return np.array(B_RSS)<jupyter_output><empty_output><jupyter_text>Now loop over a list with $b$ values between 0 and 14 with steps of 0.5. Store it in bval_RSS. Print out the resulting table.<jupyter_code>import sys b_val = np.arange(0, 14.5, .5) b_val bval_RSS = rss_values(x, y, 43, b_val) np.savetxt(sys.stdout, bval_RSS, '%16.2f') #this line is to round your result, which will make things look nicer. type(bval_RSS)<jupyter_output> 0.00 1750.97 0.50 1552.09 1.00 1368.21 1.50 1199.33 2.00 1045.45 2.50 906.57 3.00 782.69 3.50 673.81 4.00 579.93 4.50 501.05 5.00 437.17 5.50 388.29 6.00 354.41 6.50 335.53 7.00 331.65 7.50 342.77 8.00 368.89 8.50 410.01 9.00 466.13 9.50 537.25 10.00 623.37 10.50 724.49 11.00 840.61 11.50 971.73 12.00 1117.85 12.50 1278.97 13.00 1455.08 13.50 1646.20 14.00 1852.32 <jupyter_text>Plotly provides for us a table chart, and we can pass the values generated from our `rss_values` function to create a table.And let's plot this out using a a line chart.<jupyter_code>plt.figure(figsize=(10,7)) plt.plot(bval_RSS[:,0], bval_RSS[:,1], '-') plt.xlabel("b-values", fontsize=14) plt.ylabel("RSS", fontsize=14) plt.title("RSS with changes to intercept", fontsize=16);<jupyter_output><empty_output><jupyter_text>## Looking at the slope of our cost curveIn this section, we'll work up to building a gradient descent function that automatically changes our step size. To get you started, we'll provide a function called `slope_at` that calculates the slope of the cost curve at a given point on the cost curve. `Use the slope_at` function for b-values 3 and 6.<jupyter_code>def slope_at(x_values, y_values, m, b): delta = .001 base_rss = residual_sum_squares(x_values, y_values, m, b) delta_rss = residual_sum_squares(x_values, y_values, m, b + delta) numerator = delta_rss - base_rss slope = numerator/delta return {'b': b, 'slope': slope} # Use slope_at # {'b': 3, 'slope': -232.73066022784406} slope_at(x,y,43,3) # Use slope_at {'b': 6, 'slope': -52.73066022772355} slope_at(x,y,43,6) <jupyter_output><empty_output><jupyter_text>So the `slope_at` function takes in our dataset, and returns the slope of the cost curve at that point. So the numbers -232.73 and -52.73 reflect the slopes at the cost curve when b is 3 and 6 respectively.<jupyter_code>slope_3 = slope_at(x, y, 43, 3)['slope'] slope_6 = slope_at(x, y, 43, 6)['slope'] x_3 = np.linspace(3-1, 3+1, 100) x_6 = np.linspace(6-1, 6+1, 100) rss_3 = residual_sum_squares(x, y, 43, 3) rss_6 = residual_sum_squares(x, y, 43, 6) tan_3 = rss_3+slope_3*(x_3-3) tan_6 = rss_6+slope_6*(x_6-6) plt.figure(figsize=(10,7)) plt.plot(bval_RSS[:,0], bval_RSS[:,1], '-') plt.plot(x_3, tan_3, color = "red", label = "slope =" + str(round(slope_3,2))) plt.plot(x_6, tan_6, color = "green", label = "slope =" + str(round(slope_6,2))) plt.xlabel("b-values", fontsize=14) plt.ylabel("RSS", fontsize=14) plt.legend(loc='upper right', fontsize='large') plt.title("RSS with changes to slope", fontsize=16);<jupyter_output><empty_output><jupyter_text>As you can see, it seems pretty accurate. When the curve is steeper and downwards at $b = 3$, the slope is around -232.73. And at $b = 6$ with our cost curve becoming flatter, our slope is around -52.73. ## Moving towards gradient descentNow that we are familiar with our `slope_at` function and how it calculates the slope of our cost curve at a given point, we can begin to use that function with our gradient descent procedure. Remember that gradient descent works by starting at a regression line with values m, and b, which corresponds to a point on our cost curve. Then we alter our m or b value (here, the b value) by looking to the slope of the cost curve at that point. Then we look to the slope of the cost curve at the new b value to indicate the size and direction of the next step.So now let's write a function called `updated_b`. The function will tell us the step size and direction to move along our cost curve. The `updated_b` function takes as arguments an initial value of $b$, a learning rate, and the `slope` of the cost curve at that value of $m$. Its return value is the next value of `b` that it calculates.<jupyter_code>def updated_b(b, learning_rate, cost_curve_slope): return b - learning_rate * cost_curve_slope<jupyter_output><empty_output><jupyter_text>This is what our function returns.<jupyter_code>current_slope = slope_at(x, y, 43, 3)['slope'] updated_b(3, .01, current_slope) # 5.327 current_slope = slope_at(x, y, 43, 5.327)['slope'] updated_b(5.327, .01, current_slope) # 6.258 current_slope = slope_at(x, y, 43, 6.258)['slope'] updated_b(6.258, .01, current_slope) # 6.6305 current_slope = slope_at(x, y, 43, 6.631)['slope'] updated_b(6.631, .01, current_slope) # 6.780<jupyter_output><empty_output><jupyter_text>Take a careful look at how we use the `updated_b` function. By using our updated value of $b$ we are quickly converging towards an optimal value of $b$. Now let's write another function called `gradient_descent`. The inputs of the function are `x_values`, `y_values`, `steps`, the `m` we are holding constant, the `learning_rate`, and the `current_b` that we are looking at. The `steps` arguments represents the number of steps the function will take before the function stops. We can get a sense of the return value in the cell below. It is a list of dictionaries, with each dictionary having a key of the current `b` value, the `slope` of the cost curve at that `b` value, and the `rss` at that `b` value.<jupyter_code># gonna use updated b. residual sum squares. and slope at def gradient_descent(x_values, y_values, steps, current_b, learning_rate, m): result = [] for i in range(steps): bval = current_b rss = residual_sum_squares(x_values,y_values,m,current_b) slope = slope_at(x_values,y_values,m,current_b)['slope'] # print(type(slope)) # print(slope) # print(slope['slope']) result.append({'b':bval, 'rss':round(rss, 2), 'slope': round(slope, 14)}) current_b = updated_b(current_b, learning_rate, slope) return result descent_steps = gradient_descent(x, y, 50, 0, learning_rate = .005, m = 43) descent_steps #[{'b': 0, 'rss': 1750.97, 'slope': -412.73}, # {'b': 2.063653301142949, 'rss': 1026.94, 'slope': -288.91}, # {'b': 3.5082106119386935, 'rss': 672.15, 'slope': -202.24}, # {'b': 4.519400729495828, 'rss': 498.29, 'slope': -141.57}, # {'b': 5.2272338117862205, 'rss': 413.1, 'slope': -99.1}, # {'b': 5.72271696938941, 'rss': 371.35, 'slope': -69.37}, # {'b': 6.06955517971187, 'rss': 350.88, 'slope': -48.56}, # {'b': 6.312341926937677, 'rss': 340.86, 'slope': -33.99}, # {'b': 6.482292649996282, 'rss': 335.94, 'slope': -23.79}, # {'b': 6.601258156136964, 'rss': 333.53, 'slope': -16.66}, # {'b': 6.684534010435641, 'rss': 332.35, 'slope': -11.66}, # {'b': 6.742827108444089, 'rss': 331.77, 'slope': -8.16}, # {'b': 6.7836322770506285, 'rss': 331.49, 'slope': -5.71}, # {'b': 6.812195895074922, 'rss': 331.35, 'slope': -4.0}, # {'b': 6.832190427692808, 'rss': 331.28, 'slope': -2.8}]<jupyter_output><empty_output><jupyter_text>Looking at our b-values, you get a pretty good idea of how our gradient descent function works. It starts far away with $b = 0$, and the step size is relatively large, as is the slope of the cost curve. As the $b$ value updates such that it approaches a minimum of the RSS, the slope of the cost curve and the size of each step both decrease. Remember that each of these steps indicates a change in our regression line's slope value towards a "fit" that more accurately matches our dataset. Let's plot the final regression line as found before, with $m=43$ and $b=6.83$<jupyter_code>plt.scatter(x,y) plt.plot(x, 43*x+6.83)<jupyter_output><empty_output>
non_permissive
/index.ipynb
Patrickbfuller/dsc-2-14-12-gradient-descent-step-sizes-lab-seattle-ds-career-040119
12
<jupyter_start><jupyter_text># The Sparks Foundation ## #GRIPJUNE21## Submitted by MOHD AUSAAF NASIRData Science and Business Analytics Tasks Task 1 - Prediction using Supervised MLStep 1 - Import the required libraries<jupyter_code>import pandas as pd import numpy as np import matplotlib.pyplot as plt import seaborn as sns from sklearn.model_selection import train_test_split from sklearn.linear_model import LinearRegression from sklearn.metrics import mean_absolute_error<jupyter_output><empty_output><jupyter_text>Step 2 - Read data from CSV file<jupyter_code>data= pd.read_csv("http://bit.ly/w-data")<jupyter_output><empty_output><jupyter_text>Step 3 Visualize the data<jupyter_code>data.head(6)<jupyter_output><empty_output><jupyter_text>Visualize the last five scores <jupyter_code>data.tail(5) data.describe()<jupyter_output><empty_output><jupyter_text>Check for null values if any<jupyter_code>data.isnull == True data.shape sns.set_style('darkgrid') sns.scatterplot(y= data['Scores'], x= data['Hours']) plt.title('Marks Vs Study Hours', size=24) plt.ylabel('Marks Percentage', size=16) plt.xlabel('Hours Studies', size=16) plt.show() sns.regplot(x= data['Hours'], y= data['Scores']) plt.title('Regression Plot',size=24) plt.ylabel('Marks Percentage', size=16) plt.xlabel('Hours Studied', size=16) plt.show() print(data.corr())<jupyter_output> Hours Scores Hours 1.000000 0.976191 Scores 0.976191 1.000000 <jupyter_text>Step 4 - Prepare data for Machine learning algorithms and train the algorithm<jupyter_code>data.columns data.corr() X = data.iloc[:, :-1].values Y = data.iloc[:, 1].values train_X, val_X, train_Y, val_Y = train_test_split(X, Y, random_state = 0) train_X.shape, train_Y.shape, val_X.shape, val_Y.shape<jupyter_output><empty_output><jupyter_text> Step 5 - Making the predictions<jupyter_code>regression = LinearRegression() regression.fit(train_X, train_Y) print("********* Model Trained *********") pred_Y = regression.predict(val_X) prediction = pd.DataFrame({'Hours': [i[0] for i in val_X], 'Predicted Marks': [k for k in pred_Y]}) prediction<jupyter_output>********* Model Trained ********* <jupyter_text> Step 6 - Comparing Predicted Marks with Actual Marks<jupyter_code>compare_scores = pd.DataFrame({'Actual Marks': val_Y, 'Predicted Marks': pred_Y}) compare_scores<jupyter_output><empty_output><jupyter_text> Step 7 - Evaluate the model<jupyter_code>plt.scatter(x=val_X, y=val_Y, color='green') plt.plot(val_X, pred_Y, color='red') plt.title('Actual vs Predicted', size=24) plt.ylabel('Marks Percentage', size=16) plt.xlabel('Hours Studied', size=16) plt.show()<jupyter_output><empty_output><jupyter_text>Step 8 - Computing the Mean Absolute Error<jupyter_code>print('Mean absolute error: ',mean_absolute_error(val_Y,pred_Y))<jupyter_output>Mean absolute error: 4.130879918502486 <jupyter_text>Step 9 - Final Result<jupyter_code>hours = [9.25] result = regression.predict([hours]) print("Score = {}".format(round(result[0],3)))<jupyter_output>Score = 93.893
no_license
/Task 1.ipynb
AusaafNasir/Task-1---The-Sparks-Foundation
11
<jupyter_start><jupyter_text># Object Oriented Programming ## Homework Assignment #### Problem 1 Fill in the Line class methods to accept coordinates as a pair of tuples and return the slope and distance of the line.<jupyter_code>class Line: def __init__(self,coor1,coor2): self.coor1 = coor1 self.coor2 = coor2 def distance(self): return (((self.coor2[0]-self.coor1[0])**2)+((self.coor2[1]-self.coor1[1])**2))**0.5 def slope(self): return (self.coor2[1]-self.coor1[1])/(self.coor2[0]-self.coor1[0]) # EXAMPLE OUTPUT coordinate1 = (3,2) coordinate2 = (8,10) li = Line(coordinate1,coordinate2) li.distance() li.slope()<jupyter_output><empty_output><jupyter_text>________ #### Problem 2Fill in the class <jupyter_code>class Cylinder: pi = 3.14 def __init__(self,height=1,radius=1): self.height = height self.radius = radius def volume(self): return Cylinder.pi * self.height * self.radius**2 def surface_area(self): return 2 * (Cylinder.pi * self.radius**2) + 2 * Cylinder.pi * self.radius * self.height # EXAMPLE OUTPUT c = Cylinder(2,3) c.volume() c.surface_area()<jupyter_output><empty_output>
no_license
/05-Object Oriented Programming/02-Object Oriented Programming Homework.ipynb
jackwmarkham/pythonbootcamp
2
<jupyter_start><jupyter_text># Yelp Project We will be mining Yelp data today to find zip codes that have the highest rated restaurants for different categories of food. We will be using the excellent Yelp Fusion API. An API (application programming interface) is a way for developers to directly interact with a server. APIs make it easy to request specific data. The Fusion API is the third generation of Yelp APIs. There is very good documentation. ## Make a yelp account 1. Please visit the [Yelp Fusion home page](https://www.yelp.com/developers/documentation/v3) and click the button for creating an app 1. Its not important what you name your app. Just put anything down 1. After app creation you will get both a client id and a client secret. Yelp uses this to keep track of how you use their API 1. Copy and paste the client id and client secret into the strings below## Authenticate Yourself Before you can use the API, you must authenticate yourself using the standard OAuth2 protocol. You will make a post request with your id and secret and Yelp will respond with an access token. Once you have your access token you can start using the API.<jupyter_code>import pandas as pd import requests # replace CLIENT_ID and CLIENT_SECRET with your info app_id = 'CLIENT_ID' app_secret = 'CLIENT_SECRET' # don't edit these lines data = {'grant_type': 'client_credentials', 'client_id': app_id, 'client_secret': app_secret} token = requests.post('https://api.yelp.com/oauth2/token', data=data) access_token = token.json()['access_token']<jupyter_output><empty_output><jupyter_text>### Using the API The Yelp API is fairly easy to use. There are about six different **endpoints** that you can use. An endpoint is a URL that you make a web request to. Along with the endpoint you send a list of parameters that specify the results that you would like. This project will use the business search endpoint. It has a url of **https://api.yelp.com/v3/businesses/search**. There are about a dozen parameters you can use to specify an exact search. Check the [search documentation](https://www.yelp.com/developers/documentation/v3/business_search) for more detail. ### Your first search The following API call to the business search endpoint, searches for the top 50 Italian restaurants in Houston sorted by rating with price of 1, 2 or 3. The highest price is 4. This makes a web request and will take a second or so.<jupyter_code># don't edit these lines url = 'https://api.yelp.com/v3/businesses/search' headers = {'Authorization': f'bearer {access_token}'} # change these to make an API call params = {'location': 'Houston', 'categories':'italian', 'limit':'50', 'sort_by':'rating', 'price':'1,2,3' } resp = requests.get(url=url, params=params, headers=headers)<jupyter_output><empty_output><jupyter_text>### Examine response Look back at the documentation and you will see a sample response. The response for most APIs is JSON data which results directly as a Python dictionary. JSON data is a hierarchical and nested similar to how your file system stores files and directories. ### Convert response into Python dictionary The following command converts the response into a Python dictionary<jupyter_code># data is now a Python dictionary data = resp.json()<jupyter_output><empty_output><jupyter_text>### The data keys The entire response now resides **`data`** Python variable which is a dictionary. There are three keys in the dictionary.<jupyter_code>data.keys()<jupyter_output><empty_output><jupyter_text>### The businesses key The main data resides in the businesses keys. The number of results are found with **total** and the lat/long of the search region is paired with **`region`**.<jupyter_code>data['total'] data['region'] # Since data['businesses'] is a list, this will output the first 3 restaurants data['businesses'][:3]<jupyter_output><empty_output><jupyter_text>### A list of restaurants (dictionaries) By examining the output of the **`businesses`** key we see a list. This list is composed of dictionaries that contain the actual restaurant data.<jupyter_code># the business key contains a list of restaurants (dictionaries) type(data['businesses']) # we specified 50 results returned len(data['businesses'])<jupyter_output><empty_output><jupyter_text>### Look at the first restaurant<jupyter_code># Lets go to this restaurant! data['businesses'][0]<jupyter_output><empty_output><jupyter_text># Your turn It's now your turn to make a completely different request and examine the results. Use the Yelp documentation to make a different search.<jupyter_code># don't edit these lines url = 'https://api.yelp.com/v3/businesses/search' headers = {'Authorization': f'bearer {access_token}'} # CHANGE THESE PARAMETERS to make the search params = {'location': 'Houston', 'categories':'italian', 'limit':'50', 'sort_by':'rating', 'price':'1,2,3' } resp = requests.get(url=url, params=params, headers=headers) # data is now a Python dictionary that contains all the data data = resp.json()<jupyter_output><empty_output><jupyter_text>### Use the next few lines to examine the results# Project: Find the best restaurants at the best price for each zip code We will now walk through in class how to turn the JSON data into a pandas DataFrame in order to answer more interesting questions like finding the best restaurants at the best price for each zip code.<jupyter_code>with open('zip_codes.txt', 'r') as f: zip_codes = [int(line.strip()) for line in f.readlines()] # this will take a long time. You might want to use less zip codes all_restaurants = [] for zip_code in zip_codes: params = {'location': f'{zip_code}', 'categories':'restaurants', 'offset':'0', 'limit':'50' } resp = requests.get(url=url, params=params, headers=headers) cur_data = resp.json()['businesses'] all_restaurants.extend(cur_data) rows = [] for restaurant in all_restaurants: row = {} row['Category'] = restaurant['categories'][0]['title'] row['Latitude'] = restaurant['coordinates']['latitude'] row['Longitude'] = restaurant['coordinates']['longitude'] row['Phone'] = restaurant['display_phone'] row['ID'] = restaurant['id'] row['image_url'] = restaurant['image_url'] row['Address'] = restaurant['location']['address1'] row['City'] = restaurant['location']['city'] row['State'] = restaurant['location']['state'] row['Zip Code'] = restaurant['location']['zip_code'] row['Name'] = restaurant['name'] row['Price'] = restaurant.get('price', None) row['Rating'] = restaurant['rating'] row['Review Count'] = restaurant['review_count'] row['URL'] = restaurant['url'] rows.append(row) df_restaurants = pd.DataFrame(rows) df_restaurants = df_restaurants.drop_duplicates() df_restaurants.head(10)<jupyter_output><empty_output>
no_license
/Yelp API Project.ipynb
fh4/Intro-Python-Programming
9
<jupyter_start><jupyter_text>#### Merging dataframes <jupyter_code>daily2 = pd.read_csv('/Users/plarkin/Documents/GA/projects/p5/tanveer/Data/oc_daily.csv') daily2.head(1) # to load file daily = pd.read_csv('/Users/plarkin/Documents/GA/projects/p5/tanveer/clean_daily_oc.csv') daily.head(1) daily.shape,daily2.shape daily = daily.join(daily2, how ='inner',lsuffix = '_') daily.head(1) daily.columns<jupyter_output><empty_output><jupyter_text>## Sentiment Analysis#### Vader Sentiment Analysis #### Textblob Sentiment Analysis source : https://neptune.ai/blog/sentiment-analysis-python-textblob-vs-vader-vs-flair https://textblob.readthedocs.io/en/dev/<jupyter_code>#Adding in Sentiment analysis with designated columns for each output (pos, neg, neu, compound) analyzer = SentimentIntensityAnalyzer() #daily['vader'] = daily['text'].map(lambda x:analyzer.polarity_scores(str(x))) daily['vader_compound'] = [analyzer.polarity_scores(x)['compound'] for x in daily['text']] # draft_df['vd_neg'] = [analyzer.polarity_scores(x)['neg'] for x in draft_df['alltext']] # draft_df['vd_neu'] = [analyzer.polarity_scores(x)['neu'] for x in draft_df['alltext']] # draft_df['vd_pos'] = [analyzer.polarity_scores(x)['pos'] for x in draft_df['alltext']] %time from textblob import TextBlob #testimonial = TextBlob() #draft_df['tb_polarity'] = [testimonial.polarity(x)['polarity'] for x in draft_df['alltext']] #draft_df['tb_subj'] = [testimonial.sentiment(x)['subjectivity'] for x in draft_df['alltext']] daily['textblob_polarity'] = daily['text'].map(lambda words: TextBlob(str(words)).polarity) #polarity is more applicable and comparable to vader compound. subjectivity is more about opinion vs fact %time<jupyter_output>CPU times: user 2 µs, sys: 0 ns, total: 2 µs Wall time: 5.25 µs CPU times: user 2 µs, sys: 0 ns, total: 2 µs Wall time: 5.01 µs <jupyter_text>### Text Cleaning eliminate the punctuation, URL, and @ #source: https://monkeylearn.com/blog/text-cleaning/<jupyter_code>#Use this to remove http, punctuation, URL, and @ daily['text'] = daily['text'].map(lambda x: re.sub(r"(@\[A-Za-z0-9]+)|([^0-9A-Za-z \t])|(\w+:\/\/\S+)|^rt|http.+?", "", str(x.lower()))) #convert price_direction to numerical and drop first row with NA value daily.dropna(inplace=True) daily['direction'] = daily['direction'].map({'down' : -1,'same' : 0 , 'up' : 1})<jupyter_output><empty_output><jupyter_text>tokenize and lemmatize (no longer lemmatizing, results from gridsearch showed superior accuracy without lemmatizing)# Checking w/out tokenizing<jupyter_code>#for checking model with just sentiment merged_df = daily #setting date as index and dropping text column for modeling merged_df.set_index('date', inplace=True) merged_df.sort_index(inplace=True) merged_df.drop(columns= ['text', 'day_of_week','date_'], inplace=True) <jupyter_output><empty_output><jupyter_text>### Train Test Split<jupyter_code>X= merged_df.drop(columns=['Open_pct','Close_pct','Volume_diff','direction']) y=merged_df['direction'] # X = merged_df.drop(columns= ['price_direction']) # y = merged_df[['price_direction']].values # yy = merged_df['price_direction'] yy.value_counts(normalize=True) #sticking with a test size of 0.20 to save 2 years of data to test on X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42, shuffle=False, test_size=0.2) #need to one hot encode multiclass target in order to process in nn y_train = to_categorical(y_train, 3) y_test = to_categorical(y_test, 3) #source https://stackoverflow.com/questions/61550026/valueerror-shapes-none-1-and-none-3-are-incompatible # Scale for neural networks ss = StandardScaler() Xs_train = ss.fit_transform(X_train) Xs_test = ss.transform(X_test) # the length parameter dictates how many rows will constitute a sample train_sequences = TimeseriesGenerator(Xs_train, y_train, length=90, batch_size=268) #increased batch sizes from 64 to 268 # test sequences test_sequences = TimeseriesGenerator(Xs_test, y_test, length=90, batch_size=268) train_sequences[0][0].shape input_shape = train_sequences[0][0][0].shape input_shape<jupyter_output><empty_output><jupyter_text># RNN https://towardsdatascience.com/multi-class-text-classification-with-lstm-1590bee1bd17<jupyter_code># model network rnn3 = Sequential() rnn3.add(GRU(8,input_shape=input_shape, return_sequences=True)) rnn3.add(GRU(8,return_sequences=True)) rnn3.add(GRU(8,return_sequences=False)) # false if next layer dense rnn3.add(Dense(10,activation='relu')) rnn3.add(Dense(10,activation='relu')) rnn3.add(Dense(3,activation='softmax')) # compile model rnn3.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['acc']) # show summary rnn3.summary() hist_rnn3 = rnn3.fit(train_sequences,validation_data=test_sequences, epochs=100,verbose=0) # helper function to plot train and val accuracy def plot_acc(hist): plt.plot(hist.history['acc'], label='Train accuracy') plt.plot(hist.history['val_acc'], label='Test accuracy') plt.legend(); plot_acc(hist_rnn3) #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(2, activation='relu')) model.add(Dense(2, activation='relu')) model.add(Dense(3, activation='softmax')) #compile it model.compile(optimizer='Adam', loss='CategoricalCrossentropy', metrics=['acc']) #fit it #adding early stopping as a regularization technique early_stop = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='auto') history = model.fit(train_sequences, validation_data=test_sequences, epochs=100, verbose=0, callbacks=[early_stop]) #increased epochs from 100 to 300 #plot our results plt.plot(history.history['loss'], label='Train loss') plt.plot(history.history['val_loss'], label='Test loss') plt.legend(); plt.plot(history.history['acc'], label='Train accuracy') plt.plot(history.history['val_acc'], label='Test accuracy') plt.legend(); max(history.history['val_acc'])<jupyter_output><empty_output><jupyter_text>RNN Model history/specs FIRST max test accuracy 0.45553821325302124 #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(2, activation='relu')) model.add(Dense(2, activation='relu')) model.add(Dense(3, activation='softmax')) #compile it model.compile(optimizer=Adam(learning_rate=.0005), loss='CategoricalCrossentropy', metrics=['acc']) #fit it history = model.fit(train_sequences, validation_data=test_sequences, epochs=100, verbose=0) notes: maxing out after 15-20 epochs SECOND max test accuracy 0.4570982754230499 #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(8, activation='relu'))#added and increased both hidden layers to 8 model.add(Dense(8, activation='relu')) model.add(Dense(3, activation='softmax')) #compile it model.compile(optimizer=Adam(learning_rate=.0005), loss='CategoricalCrossentropy', metrics=['acc']) #fit it history = model.fit(train_sequences, validation_data=test_sequences, epochs=300, verbose=0) #increased epochs from 100 to 300 notes: maxing out after 20 epochs THIRD 0.4586583375930786 #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(32, activation='relu'))#added and increased both hidden layers to 32 model.add(Dense(32, activation='relu')) model.add(Dense(3, activation='softmax')) #compile it model.compile(optimizer='Adam', loss='CategoricalCrossentropy', metrics=['acc']) #fit it history = model.fit(train_sequences, validation_data=test_sequences, epochs=20, verbose=0) #decreased epochs from 300 to 20 FOURTH 0.4602183997631073 added earlystopping #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(32, activation='relu'))#added and increased both hidden layers to 32 model.add(Dense(32, activation='relu')) model.add(Dense(3, activation='softmax')) #softmax for multi-classification #compile it model.compile(optimizer='Adam', loss='CategoricalCrossentropy', metrics=['acc']) #categorical crossentropy for multi-classification #fit it #adding early stopping as a regularization technique early_stop = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='auto') history = model.fit(train_sequences, validation_data=test_sequences, epochs=20, verbose=0, callbacks=[early_stop]) FIFTH 0.4492979645729065 #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(32, activation='relu'))#added and increased both hidden layers to 32 model.add(BatchNormalization()) #added to help regularize model model.add(Dense(32, activation='relu')) model.add(Dense(3, activation='softmax')) #softmax for multi-classification #compile it model.compile(optimizer='Adam', loss='CategoricalCrossentropy', metrics=['acc']) #categorical crossentropy for multi-classification #fit it #adding early stopping as a regularization technique early_stop = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='auto') history = model.fit(train_sequences, validation_data=test_sequences, epochs=20, verbose=0, callbacks=[early_stop]) #increased epochs from 100 to 300 SIXTH 0.4836193323135376 adding dropout 0.2 #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(32, activation='relu'))#added and increased both hidden layers to 32 model.add(BatchNormalization()) #added to help regularize model model.add(Dropout(0.2)) #added to help regularize model model.add(Dense(32, activation='relu')) model.add(Dropout(0.2)) #added to help regularize model model.add(Dense(3, activation='softmax')) #softmax for multi-classification #compile it model.compile(optimizer='Adam', loss='CategoricalCrossentropy', metrics=['acc']) #categorical crossentropy for multi-classification #fit it #adding early stopping as a regularization technique early_stop = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='auto') history = model.fit(train_sequences, validation_data=test_sequences, epochs=20, verbose=0, callbacks=[early_stop]) #increased epochs from 100 to 300 SEVENTH 0.4492979645729065 adding kernel regularization l1 #modeling and layers model = Sequential() model.add(GRU(8, input_shape=input_shape, return_sequences= True)) model.add(GRU(8, return_sequences= False)) model.add(Dense(32, activation='relu'))#added and increased both hidden layers to 32 model.add(BatchNormalization()) #added to help regularize model model.add(Dropout(0.2)) #added to help regularize model model.add(Dense(32, activation='relu', kernel_regularizer='l1')) model.add(Dropout(0.2)) #added to help regularize model model.add(Dense(3, activation='softmax', kernel_regularizer='l1')) #softmax for multi-classification #compile it model.compile(optimizer='Adam', loss='CategoricalCrossentropy', metrics=['acc']) #categorical crossentropy for multi-classification #fit it #adding early stopping as a regularization technique early_stop = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='auto') history = model.fit(train_sequences, validation_data=test_sequences, epochs=20, verbose=0, callbacks=[early_stop]) #increased epochs from 100 to 300 # LSTM<jupyter_code># make df? X_train_lstm= np.reshape(Xs_train,(Xs_train.shape[0],1,Xs_train.shape[1])) X_test_lstm = np.reshape(Xs_test,(Xs_test.shape[0],1,Xs_test.shape[1])) X_train_lstm.shape, X_test_lstm.shape # model network lstm = Sequential() lstm.add(LSTM(64,input_shape=(1,16))) lstm.add(Dense(8,activation='relu')) lstm.add(Dense(3,activation='softmax')) # compile model lstm.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['acc']) lstm.summary() # fit model hist_lstm = lstm.fit(X_train_lstm,y_train, validation_data=(X_test_lstm,y_test), epochs=100,verbose=0) plot_acc(hist_lstm) max(hist_lstm.history['val_acc'])<jupyter_output><empty_output>
no_license
/classification_with_text/RNN & LSTM Modeling-Only Sentiment.ipynb
plarkin13/stock_prediction_sentiment
7
<jupyter_start><jupyter_text># Predicting eye tracking features Eye-tracking data from reading represent an important resource for both linguistics and natural language processing. The ability to accurately model gaze features is crucial to advance our understanding of language processing. On one hand, it can tell us how well our computational models align with human language processing mechanisms. On the other hand, it may provide more accurate models of reading to further psycholinguistic research. In this tutorial we will first get acquainted with eye tracking data and the structure of the [Zurich Cognitive Language Processing Corpus (ZuCo)](https://www.nature.com/articles/sdata2018291), for which eye movements were recorded during natural reading of English sentences. We use the eye tracking data as is was provided for the [CMCL 2021 Shared Task](https://cmclorg.github.io/shared_task). Then, we evaluate the ability of a contextualized language model to predict psycholinguistic data, specifically we use transformer models such as BERT and DistilBERT to predict eye tracking features recorded during reading (fixation duration, fixation proportion, etc.). The goal of the task is to predict 5 different token-level eye-tracking metrics from the ZuCo corpus. Finally, we compare the performance of a pre-trained model and a model fine-tuned on eye tracking data.## Prerequisites<jupyter_code># Install the required pip packages in the current Jupyter kernel import sys !{sys.executable} -m pip install numpy pandas seaborn matplotlib torch transformers spacy # Import the required libraries import numpy as np import pandas as pd import random import torch import transformers import seaborn as sns import matplotlib.pyplot as plt<jupyter_output><empty_output><jupyter_text>## Loading dataset<jupyter_code># If you are workin on Google Colab, you need to clone the repo to get access to the data files: #!git clone https://github.com/beinborn/ESSLLI2021.git # You still need to make sure that the files are under the correct paths! # Load ZuCo corpus (train, test and gold data files) training_data = pd.read_csv("data/training_data.csv") gold_data = pd.read_csv("data/gold_data.csv") print(len(training_data['sentence_id'].unique()), "training sentences including", len(training_data), "words.") print(len(gold_data['sentence_id'].unique()), "test sentences including", len(gold_data), "words.")<jupyter_output>800 training sentences including 15736 words. 191 test sentences including 3554 words. <jupyter_text>## Data analysis Originally, the ZuCo corpus contains eye tracking data from 30 participants. Here, the data has already been averaged across readers and the feature values have been scaled to a range between 0 and 100. The eye-tracking feature values are scaled to facilitate evaluation via the mean absolute error. The features nFix and fixProp are scaled separately, while FFD, GPT and TRT are scaled together since these are all dependent and measured in milliseconds. The data was randomly shuffled before splitting into training and test data.<jupyter_code># Let's have a look at some sentences and their eye tracking features to get a better idea of the data. training_data.loc[training_data['sentence_id'] == 444] # Note: change the sentence_id to see other sentences (0 - 799)<jupyter_output><empty_output><jupyter_text>### Eye tracking features explained For each word, there are five eye-tracking features: - _nFix_ (number of fixations): total number of fixations on the current word. - _FFD_ (first fixation duration): the duration of the first fixation on the prevailing word. - _GPT_ (go-past time): the sum of all fixations prior to progressing to the right of the current word, including regressions to previous words that originated from the current word. - _TRT_ (total reading time): the sum of all fixation durations on the current word, including regressions. - _fixProp_ (fixation proportion): the proportion of participants that fixated the current word (as a proxy for how likely a word is to be fixated).#### A note on tokenization: The tokens in the sentences are split in the same manner as they were presented to the participants during the eye tracking experiments. Hence, this does not necessarily follow a linguistically correct tokenization. For example, the sequences ''(except,'' and ''don't'' were presented as such to the reader and not split into ''('', ''except'', '','' and ''do'', ''n't'' as a tokenizer would do to ensure a more natural way of reading. Sentence endings are marked with an \ symbol added to the last token. To get a better idea of the data, let's plot some sentences and their corresponding eye tracking features...<jupyter_code># Visualization: fixation times # Note: change the sentence_id to see other sentences (0 - 799) example_sentence = training_data.loc[training_data['sentence_id'] == 444] fig = plt.figure(figsize=(len(example_sentence)+2,3)) feature_to_draw = "TRT" # or "GPT" or "TRT" words = [s.replace("<EOS>", "") for s in example_sentence["word"]] ax = sns.scatterplot(data=example_sentence, x="word_id", y=0, size=feature_to_draw, hue=feature_to_draw, palette="flare", sizes=(200,5000), legend=False) ax.set_frame_on(False) ax.tick_params(axis=u'both', which=u'both',length=0) plt.xticks(ticks=example_sentence["word_id"], labels=words, fontsize=14) plt.yticks([0], [feature_to_draw], fontsize=14) plt.ylim(-0.05,0.07) plt.xlabel(None) plt.tight_layout() plt.show() # Visualization: fixation proportion # Note: change the sentence_id to see other sentences (0 - 799) example_sentence = training_data.loc[training_data['sentence_id'] == 444] fig = plt.figure(figsize=(len(example_sentence)+2,3)) feature_to_draw = "fixProp" # or nFix words = [s.replace("<EOS>", "") for s in example_sentence["word"]] ax = sns.barplot(data=example_sentence, x="word_id", y="fixProp", color="lightblue") ax.set_frame_on(False) ax.tick_params(axis=u'both', which=u'both',length=0) plt.xticks(ticks=example_sentence["word_id"], labels=words, fontsize=14) plt.xlabel(None) plt.ylabel(feature_to_draw, fontsize=16) plt.tight_layout() plt.show() # Now let's get a complete overview of the training data c = sns.color_palette("hls", 10) sns.set(font_scale = 1) sns.set_style("whitegrid") ax = sns.boxplot(data=training_data[["nFix","FFD","GPT","TRT","fixProp"]], palette=c[5:], color='grey', linewidth=1, fliersize=1) medians = [] for f in ["nFix","FFD","GPT","TRT","fixProp"]: median = training_data[f].median() medians.append(median) median_labels = [str(np.round(s, 2)) for s in medians] pos = range(len(medians)) for tick,label in zip(pos,ax.get_xticklabels()): ax.text(pos[tick], -4, median_labels[tick], #medians[tick] + offsets[tick] horizontalalignment='center', size='small', color='black')#, weight='semibold') plt.show() <jupyter_output><empty_output><jupyter_text>## Training model We compare three state-of-the-art language models trained for English: BERT and DistilBERT. BERT was the first widely successful transformer-based language model and remains highly influential. DistilBERT is a variant of BERT that requires less training time due to a considerable reduction of the training parameters while maintaining similar performance on benchmark datasets. This allows us to analyze if the lighter architectures have an influence on the patterns of eye movements that the models learn.<jupyter_code># import source code for training and evaluating models import model import dataloader import evaluation_metric<jupyter_output><empty_output><jupyter_text>Training the model can take a while. It's best to start with a low number of epochs. Alternatively, you can skip the next steps and use the provided models in the "models/" directory: - A pre-trained DistilBERT model (i.e., fine-tuned for 0 epochs) - A fine-tuned DistilBERT model (trained for 150 epochs on the eye-tracking data)<jupyter_code># Note: # You could load any other BERT or DistilBERT model version available in the Huggingface library in the same manner. # DistilBERT was chosen for this tutorial since it has less parameters. random.seed(12345) device = torch.device('cpu') # define model model_name = 'distilbert-base-uncased' regr_model = model.TransformerRegressionModel(model_name).to(device) train_data = dataloader.EyeTrackingCSV(training_data, model_name=model_name) train_loader = torch.utils.data.DataLoader(train_data, batch_size=16, shuffle=True) optimizer = torch.optim.AdamW(regr_model.parameters(), lr=5e-5) loss = torch.nn.MSELoss() # train model # Note: start with 1 or 2 epochs, so find out how long this takes... # set to 0 or empty list for no fine-tuning? num_epochs = 5 # is this needed? print("Starting training...") for epoch in range(num_epochs): model.train(regr_model, train_loader, optimizer, loss) print("Epoch:", epoch) print("Training done.") # save model output_model = './models/'+model_name+'_'+str(num_epochs) torch.save(regr_model, output_model) print("Model saved to", output_model)<jupyter_output>Some weights of the model checkpoint at distilbert-base-uncased were not used when initializing DistilBertModel: ['vocab_layer_norm.weight', 'vocab_transform.bias', 'vocab_projector.bias', 'vocab_layer_norm.bias', 'vocab_projector.weight', 'vocab_transform.weight'] - This IS expected if you are initializing DistilBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model). - This IS NOT expected if you are initializing DistilBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model). <jupyter_text>Note: You can ignore the warning about the model checkpoint... ## Evaluate modelsLoad pre-trained model and predict eye tracking features <jupyter_code># load pretrained model pretrained_model_to_evaluate = "models/distilbert-base-uncased_0" pretrained_model = torch.load(pretrained_model_to_evaluate) predictions_pretrained = model.predict(pretrained_model, pretrained_model_to_evaluate, gold_data) predictions_pretrained # load model fine-tuned on eye-tracking data finetuned_model_to_evaluate = "models/distilbert-base-uncased_150" finetuned_model = torch.load(finetuned_model_to_evaluate) predictions_finetuned = model.predict(finetuned_model, finetuned_model_to_evaluate, gold_data) predictions_finetuned<jupyter_output><empty_output><jupyter_text>### Baseline We use the mean central tendency as a baseline for this regression problem, i.e., we calculate the mean value for each feature from the training data and use it as a prediction for all words in the test data.<jupyter_code># compute mean values from training file features = ["nFix","FFD","GPT","TRT","fixProp"] mean_feat_values = {} print("Mean feature values in training data:") for feat in features: avg = np.mean(training_data[feat]) mean_feat_values[feat] = avg print (feat, avg) # make dataframe of test data with mean values mean_data = gold_data.copy() for feat in features: mean_data[feat] = mean_feat_values[feat]<jupyter_output>Mean feature values in training data: nFix 15.1017782748365 FFD 3.1910228250548047 GPT 6.345511205701235 TRT 5.309609916601204 fixProp 67.05723394392223 <jupyter_text>### Evaluation metric: Mean absolute error Now that we have the predictions of the mean baseline as well as a pre-trained and a fine-tuned model, we can calculate the resutss. The model performance is evaluated by using the _mean absolute error_ metric (MAE), meaning that we average over the error of all samples in the test set: $MAE = \frac{\sum_{i=1}^{n} | y_{i} - x_{i}|}{n} $ <jupyter_code># Calculate mean absolute error on the predictions print("Results pre-trained model:") mae_pretrained = evaluation_metric.evaluate(predictions_pretrained, gold_data) print("Results fine-tuned model:") mae_finetuned = evaluation_metric.evaluate(predictions_finetuned, gold_data) print("Results mean baseline:") mae_mean = evaluation_metric.evaluate(mean_data, gold_data) data = pd.DataFrame(data={'results': ["mean", "pretrained", "finetuned"], 'mae': [mae_mean, mae_pretrained, mae_finetuned]}) ax = sns.barplot(x="results", y="mae", data=data, palette="flare") ax.set_ylabel('MAE overall', fontsize=16) plt.show()<jupyter_output>Results pre-trained model: MAE for nFix: 15.087638004183914 MAE for FFD: 2.9985058935006013 MAE for GPT: 6.3720704862735795 MAE for TRT: 5.256835941134088 MAE for fixProp: 67.01796054929119 Overall MAE: 19.346602174876672 Results fine-tuned model: MAE for nFix: 4.3378435708878325 MAE for FFD: 0.7026542748697112 MAE for GPT: 2.5579450948635043 MAE for TRT: 1.6773681869889472 MAE for fixProp: 11.880517177967716 Overall MAE: 4.2312656611155415 Results mean baseline: MAE for nFix: 7.302784572412885 MAE for FFD: 1.1491256623904909 MAE for GPT: 3.781862596034746 MAE for TRT: 2.777713539662726 MAE for fixProp: 21.77547317917012 Overall MAE: 7.357391909934194 <jupyter_text>As the numbers and the plot show, the pre-trained model that is not trained on eye-tracking features but purely on text, does not beat the mean baseline. However, the model fine-tuned on eye-tracking data is substantially better (i.e., lower MAE) than the baseline. Based on these results, we can conclude that the difficulty of predicting the individual eye-tracking features is analogous in all models: FFD is the most accurately predicted feature. This seems to suggest that the models are more capable to capture early processing stages of lexical access compared to late-stage semantic integration, represented by TRT and nFix. Generally, the error for the three features representing reading times in milliseconds (FFD, GPT, and TRT), is much lower than for nFix and fixProp. The latter are the features with the most variance. The mean baseline results also reveal the same patterns. The features with lower variance achieve lower MAEs. The fixProp feature, representing how likely a word is to be fixated, might be more challenging to predict since it is more dependent on subject-specific characteristics. ### Visualization of predictions<jupyter_code># plot a sentence with real eye tracking data and the corresponding model predictions sentence_to_plot = 953 #801, 901, 950, 953 feature_to_plot = "FFD" true_sentence = gold_data.loc[gold_data['sentence_id'] == sentence_to_plot] predicted_sentence_finetuned = predictions_finetuned.loc[predictions_finetuned['sentence_id'] == sentence_to_plot] predicted_sentence_pretrained = predictions_pretrained.loc[predictions_pretrained['sentence_id'] == sentence_to_plot] # todo: change fig size accroding to sent length fig, ax = plt.subplots(1,1,figsize=(len(true_sentence)*1.5,5)) ax.plot(list(range(len(true_sentence))), true_sentence[feature_to_plot], label="correct", color="blue", linestyle="--") ax.plot(list(range(len(predicted_sentence_finetuned))), predicted_sentence_finetuned[feature_to_plot], label="fine-tuned", color="red", marker='o', ms=10, linestyle="-.") ax.plot(list(range(len(predicted_sentence_pretrained))), predicted_sentence_pretrained[feature_to_plot], label="pretrained", color="pink", marker='o', ms=10, linestyle="-.") # todo: add mean as a task ax.plot(list(range(len(predicted_sentence_pretrained))), [mean_feat_values[feature_to_plot]]*len(predicted_sentence_pretrained), label="mean", color="lightblue", marker='o', ms=10, linestyle="--") ax.set_ylabel(feature_to_plot, fontsize=16) ax.set_xticks(list(range(len(true_sentence)))) ax.set_xticklabels(true_sentence['word'], fontsize=16) ax.legend(fontsize=13) plt.show()<jupyter_output><empty_output><jupyter_text>### Part of Speech Analysis Now we have both evaluated the aggregated performance of the models as well as looked at the predictions for specific sentences. Next, we want to analyze whether there are certain parts of speech (PoS) for which eye tracking features are more easily predictable than others. Therefore, we first use the SpaCy part-of-speech tagger to get the PoS tags for the sentence in our test data: <jupyter_code># load English SpaCy model import spacy !python3 -m spacy download en_core_web_sm nlp = spacy.load("en_core_web_sm") # making sure Spacy doesn't mess with out tokenization from spacy.tokens import Doc def custom_tokenizer(wordlist): """replace spacy tokenizer with already tokenized list of words""" return Doc(nlp.vocab, words=wordlist, spaces=None) nlp.tokenizer = custom_tokenizer # tag the data and add the tags to the dataframe def pos_tag(data_to_tag): words = [str(w).replace("<EOS>", ".") for w in data_to_tag['word']] tagged_sents = nlp(words) data_to_tag['pos_tags'] = [token.pos_ for token in tagged_sents] return data_to_tag tagged_data = pos_tag(gold_data) tagged_predictions = pos_tag(predictions_finetuned) feature_to_plot = "TRT" #fig = plt.figure(figsize=(10,5)) fig, ax = plt.subplots(1,1,figsize=(10,5)) sns.barplot(x="pos_tags", y=feature_to_plot, data=tagged_data, color="green", label="true eye tracking data", alpha=0.4) sns.barplot(x="pos_tags", y=feature_to_plot, data=tagged_predictions, color="red", label="predicted", alpha=0.4) plt.legend() plt.show()<jupyter_output><empty_output>
no_license
/code/tutorial1/esslli_tutorial1.ipynb
yancong222/ESSLLI2021
11
<jupyter_start><jupyter_text>### Control Statements - Conditional Statements - if-else - Iterative Statements - for - while - do-while<jupyter_code>year = int(input("Enter a year ")) if year%400 == 0 or (year% 100 !=0 and year%4 ==0): print("Leap Year") else: print("Non-Leap Year") n=int(input("Enter a number ")) i=1 while(i<=n): print(i,end=" ") i=i+1 n=int(input("Enter a number ")) i=1 s=0 while(i<=n): if(i%2==0): s=s+i i=i+1 print("The Sum of even numbers from 1 to",n,"is",s) n=int(input("Enter a number ")) i=0 while(n!=0): i=n%10 print(i,end=" ") n=n//10 # using / will not give us the required answer. Instead it will divide till the last of the decimal points.<jupyter_output>Enter a number 23 3 2 <jupyter_text>### Functional Programming - Simple - Better Reusability - Easy to Understand - Division of Program into small blocks<jupyter_code>def fun1(): n=int(input("Enter a number ")) i=0 s=0 while(n!=0): i=n%10 if(i%2==0): s=s+i n=n//10 return s fun1() def fun1(): n=int(input("Enter a number ")) i=0 d=0 while(n!=0): i=n%10 if(i>d): d=i n=n//10 return d fun1() def fun1(): n=int(input("Enter a number ")) i=0 s=0 p=n while(n!=0): f=1 i=n%10 while(i!=0): f=f*i i=i-1 s=s+f n=n//10 if(p==s): print("Yes") else: print("No") fun1() def fun1(): n=int(input("Enter a number ")) d=n p=0 r=0 while(n!=0): r=n%10 p=p*10 + r n=n//10 if(d==p): print("Palindrome ") else: print("Not Palindrome") fun1()<jupyter_output>Enter a number 123 Not Palindrome <jupyter_text>### When to use which loop - For non-fixed number of iterations,use the while loop - For a fixed number of iterations,use the for loop<jupyter_code>def printSeries(lb,ub): for x in range(lb,ub): # We know that range function is as follows : range(a,b-1) print(x,end=" ") return printSeries(11,25) def printAlternate(lb,ub): for x in range(lb,ub+1,2): print(x,end=" ") return lb=int(input("Enter the lower bound ")) ub=int(input("Enter the upper bound ")) printAlternate(lb,ub) def factors(): n=int(input("Enter a number ")) i=1 while(i<=n): if(n%i==0): print(i,end=" ") i=i+1 factors() def prime(): n=int(input("Enter a number ")) i=2 c=0 while(i<n): if(n%i==0): c=c+1 i=i+1 if(c==0): print("True") else: print("False ") prime() def checkprime(i): j=2 c=0 while(j<i): if(i%j==0): c=c+1 i=i+1 if(c==0): return 1 else: return 0 def primec(): n=int(input("Enter a number ")) i=2 pc=0 while(i<=n): pc=pc+checkprime(i) i=i+1 print(pc) primec()<jupyter_output><empty_output>
no_license
/Workshop_File--Day 2.ipynb
Amogh19/Problem-Solving-and-Programming
3
<jupyter_start><jupyter_text># Exercise 2: Creating Redshift Cluster using the AWS python SDK ## An example of Infrastructure-as-code<jupyter_code>import pandas as pd import boto3 import json<jupyter_output><empty_output><jupyter_text># STEP 0: Make sure you have an AWS secret and access key - Create a new IAM user in your AWS account - Give it `AdministratorAccess`, From `Attach existing policies directly` Tab - Take note of the access key and secret - Edit the file `dwh.cfg` in the same folder as this notebook and fill [AWS] KEY= YOUR_AWS_KEY SECRET= YOUR_AWS_SECRET # Load DWH Params from a file<jupyter_code>import configparser config = configparser.ConfigParser() config.read_file(open('dwh.cfg')) KEY = config.get('AWS','KEY') SECRET = config.get('AWS','SECRET') DWH_CLUSTER_TYPE = config.get("DWH","DWH_CLUSTER_TYPE") DWH_NUM_NODES = config.get("DWH","DWH_NUM_NODES") DWH_NODE_TYPE = config.get("DWH","DWH_NODE_TYPE") DWH_CLUSTER_IDENTIFIER = config.get("DWH","DWH_CLUSTER_IDENTIFIER") DWH_DB = config.get("DWH","DWH_DB") DWH_DB_USER = config.get("DWH","DWH_DB_USER") DWH_DB_PASSWORD = config.get("DWH","DWH_DB_PASSWORD") DWH_PORT = config.get("DWH","DWH_PORT") DWH_IAM_ROLE_NAME = config.get("DWH", "DWH_IAM_ROLE_NAME") (DWH_DB_USER, DWH_DB_PASSWORD, DWH_DB) pd.DataFrame({"Param": ["DWH_CLUSTER_TYPE", "DWH_NUM_NODES", "DWH_NODE_TYPE", "DWH_CLUSTER_IDENTIFIER", "DWH_DB", "DWH_DB_USER", "DWH_DB_PASSWORD", "DWH_PORT", "DWH_IAM_ROLE_NAME"], "Value": [DWH_CLUSTER_TYPE, DWH_NUM_NODES, DWH_NODE_TYPE, DWH_CLUSTER_IDENTIFIER, DWH_DB, DWH_DB_USER, DWH_DB_PASSWORD, DWH_PORT, DWH_IAM_ROLE_NAME] })<jupyter_output><empty_output><jupyter_text>## Create clients for EC2, S3, IAM, and Redshift<jupyter_code>ec2 = boto3.client( "ec2", region_name="us-west-2", aws_access_key_id=KEY, aws_secret_access_key=SECRET ) s3 = boto3.client( "s3", region_name="us-west-2", aws_access_key_id=KEY, aws_secret_access_key=SECRET ) iam = boto3.client( "iam", region_name="us-west-2", aws_access_key_id=KEY, aws_secret_access_key=SECRET ) redshift = boto3.client( "redshift", region_name="us-west-2", aws_access_key_id=KEY, aws_secret_access_key=SECRET )<jupyter_output><empty_output><jupyter_text>## Check out the sample data sources on S3<jupyter_code>for obj in s3.list_objects(Bucket="awssampledbuswest2", Prefix="ssbgz/")["Contents"]: print(obj["Key"])<jupyter_output>ssbgz/ ssbgz/customer0002_part_00.gz ssbgz/dwdate.tbl.gz ssbgz/lineorder0000_part_00.gz ssbgz/lineorder0001_part_00.gz ssbgz/lineorder0002_part_00.gz ssbgz/lineorder0003_part_00.gz ssbgz/lineorder0004_part_00.gz ssbgz/lineorder0005_part_00.gz ssbgz/lineorder0006_part_00.gz ssbgz/lineorder0007_part_00.gz ssbgz/part0000_part_00.gz ssbgz/part0001_part_00.gz ssbgz/part0002_part_00.gz ssbgz/part0003_part_00.gz ssbgz/supplier.tbl_0000_part_00.gz ssbgz/supplier0001_part_00.gz ssbgz/supplier0002_part_00.gz ssbgz/supplier0003_part_00.gz <jupyter_text>## STEP 1: IAM ROLE - Create an IAM Role that makes Redshift able to access S3 bucket (ReadOnly)<jupyter_code>assume_role_doc = json.dumps({ "Version": "2012-10-17", "Statement": [ { "Effect": "Allow", "Principal": { "Service": "redshift.amazonaws.com" }, "Action": "sts:AssumeRole" } ] }) dwhRole = iam.create_role( RoleName=DWH_IAM_ROLE_NAME, Description="Allow redshift to read from S3 Bucket", AssumeRolePolicyDocument=assume_role_doc ) iam.attach_role_policy( PolicyArn="arn:aws:iam::aws:policy/AmazonS3ReadOnlyAccess", RoleName=DWH_IAM_ROLE_NAME ) # Copied from AWS Console ## https://console.aws.amazon.com/iam/home?region=us-west-1#/policies/arn:aws:iam::aws:policy/AmazonS3ReadOnlyAccess$serviceLevelSummary ## arn:aws:iam::aws:policy/AmazonS3ReadOnlyAccess roleArn = iam.get_role( RoleName=DWH_IAM_ROLE_NAME )["Role"]["Arn"]<jupyter_output><empty_output><jupyter_text>## STEP 2: Redshift Cluster - Create a RedShift Cluster - For complete arguments to `create_cluster`, see [docs](https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/redshift.html#Redshift.Client.create_cluster)<jupyter_code>response = redshift.create_cluster( # add parameters for hardware DBName=DWH_DB, ClusterIdentifier=DWH_CLUSTER_IDENTIFIER, ClusterType=DWH_CLUSTER_TYPE, NodeType=DWH_NODE_TYPE, NumberOfNodes=int(DWH_NUM_NODES), # add parameters for identifiers & credentials MasterUsername=DWH_DB_USER, MasterUserPassword=DWH_DB_PASSWORD, # add parameter for role (to allow s3 access) IamRoles=[ roleArn ] )<jupyter_output><empty_output><jupyter_text>## 2.1 *Describe* the cluster to see its status - run this block several times until the cluster status becomes `Available`<jupyter_code>metadata_of_interest = [ "ClusterIdentifier", "NodeType", "ClusterStatus", "MasterUsername", "DBName", "Endpoint", "NumberOfNodes", "VpcId" ] describe_dict = redshift.describe_clusters( ClusterIdentifier=DWH_CLUSTER_IDENTIFIER )["Clusters"][0] {k: v for (k, v) in describe_dict.items() if k in metadata_of_interest} print("DWH_ENDPOINT: ", describe_dict["Endpoint"]["Address"]) print("DWH_ROLE_ARN: ", describe_dict["IamRoles"][0]["IamRoleArn"]) print("Role ARN: ", roleArn) DWH_ENDPOINT = describe_dict["Endpoint"]["Address"] DWH_ROLE_ARN = describe_dict["IamRoles"][0]["IamRoleArn"]<jupyter_output>DWH_ENDPOINT: dwhcluster.ctcwpuiwdyvj.us-west-2.redshift.amazonaws.com DWH_ROLE_ARN: arn:aws:iam::702717750718:role/dwhRole Role ARN: arn:aws:iam::702717750718:role/dwhRole <jupyter_text> 2.2 Take note of the cluster endpoint and role ARN DO NOT RUN THIS unless the cluster status becomes "Available" ## STEP 3: Open an incoming TCP port to access the cluster ednpoint<jupyter_code>defaultSg = ec2.describe_security_groups( GroupNames=["default"] )["SecurityGroups"][0]["GroupId"] ## https://us-west-1.console.aws.amazon.com/ec2/home?region=us-west-1#SecurityGroup:group-id=sg-2d57a950 responseSg = ec2.authorize_security_group_ingress( IpProtocol="TCP", CidrIp="0.0.0.0/0", FromPort=int(DWH_PORT), ToPort=int(DWH_PORT), GroupId=defaultSg # GroupName="defaultSg" )<jupyter_output><empty_output><jupyter_text>## STEP 4: Make sure you can connect to the clusterConnect to the cluster<jupyter_code>%load_ext sql conn_string="postgresql://{}:{}@{}:{}/{}".format(DWH_DB_USER, DWH_DB_PASSWORD, DWH_ENDPOINT, DWH_PORT,DWH_DB) print(conn_string) %sql $conn_string %%sql SELECT * FROM information_schema.tables WHERE table_schema = 'public'<jupyter_output> * postgresql://dwhuser:***@dwhcluster.ctcwpuiwdyvj.us-west-2.redshift.amazonaws.com:5439/dwh 7 rows affected. <jupyter_text>## STEP 5: Clean up your resourcesDO NOT RUN THIS UNLESS YOU ARE SURE We will be using these resources in the next exercises<jupyter_code>#### CAREFUL!! #-- Uncomment & run to delete the created resources redshift.delete_cluster( ClusterIdentifier=DWH_CLUSTER_IDENTIFIER, SkipFinalClusterSnapshot=True) #### CAREFUL!!<jupyter_output><empty_output><jupyter_text>- run this block several times until the cluster really deleted<jupyter_code>{k: v for (k, v) in describe_dict.items() if k in metadata_of_interest} #### CAREFUL!! #-- Uncomment & run to delete the created resources iam.detach_role_policy(RoleName=DWH_IAM_ROLE_NAME, PolicyArn="arn:aws:iam::aws:policy/AmazonS3ReadOnlyAccess") iam.delete_role(RoleName=DWH_IAM_ROLE_NAME) #### CAREFUL!!<jupyter_output><empty_output>
no_license
/cloud_data_warehouses/.ipynb_checkpoints/L3 Exercise 2-checkpoint.ipynb
aaharrison/udacity-data-eng
11
<jupyter_start><jupyter_text># 50 Years of Music Trends ## Objective * Analyze lyrics from billboard top 100 songs over 50 years to identify trends * Statement: Has the sentiments of popular lyrics changed over time? ## Hypothesis * Ha = the sentiments of popular lyrics has become more negative over time * Ho = no change in the sentiments of popular lyrics over time ## Sources * musixmatch source: https://developer.musixmatch.com/documentation/api-reference/track-lyrics-get * musixmatch python: https://github.com/hudsonbrendon/python-musixmatch * billboard python: https://github.com/guoguo12/billboard-charts<jupyter_code># Dependency library import numpy as np import pandas as pd import random from datetime import datetime, timedelta import matplotlib.pyplot as plt from scipy.stats import linregress # API Calls import billboard from musixmatch import Musixmatch # API Keys from musixmatch_api import api_key # Import and Initialize Sentiment Analyzer from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer analyzer = SentimentIntensityAnalyzer() # Generate a (pseudo) random list of (almost all) dates in string format to fit musixmatch parameter date_list = [] min_year = 1968 error_counter = 0 for i in range(50): try: # RANDOM date generation month_rand = str(random.randint(1,12)) if len(month_rand) ==1: # PAD single digit numbers with a leading 0 month_rand = month_rand.zfill(2) day_rand = str(random.randint(1,28)) if len(day_rand) == 1: day_rand = day_rand.zfill(2) # STRINGIFY the result for the musixmatch parameter date_rand = str(f'{min_year}-{month_rand}-{day_rand}') # APPEND result to date_list date_list.append(date_rand) #INCREMENT the year min_year = min_year + 1 except ValueError: error_counter = error_counter + 1 # VIEW date_list object date_list # Return top 100 billboard songs for each date in random list generated above # Note: Running this code takes approximately 2 minutes billboard_list = 'hot-100' col_names = ['Song','Artist','Date'] chart_df = pd.DataFrame(columns = col_names) for date in date_list: chart = billboard.ChartData(billboard_list,date) for row in chart: # EMPTY the list placeholder used to create the dataframe chart_list = [] # CLEAN and convert the billboard object chart_list.append(str(row).replace("'","",1)) temp_df = pd.DataFrame(chart_list) temp_df = temp_df[0].str.split("' by ",expand=True) temp_df = temp_df.rename(columns={0:"Song",1:"Artist"}) temp_df['Date'] = date # APPEND the temp_df in the current loop location to the chart_df chart_df = chart_df.append(temp_df) # REMOVE duplicates and RESET index from the resulting dataframe chart_df = chart_df.drop_duplicates().reset_index(drop=True) # VIEW dataframe head print(len(chart_df)) chart_df.head() # SPLIT the date values in the dataframe for plotting and analysis purposes chart_df['Year'], chart_df['Month'], chart_df['Day'] = chart_df['Date'].str.split('-').str # VIEW dataframe head chart_df.head() # Retrieve lyrics from MusixMatch API based on song and artist in above dataframe # Running this code takes approximately 5 - 7 minutes musixmatch = Musixmatch(api_key) lyrics_list = [] error_counter = 0 # LOOP through the data frame and use song title and artist name to search for lyrics in musixmatch for x in range(len(chart_df)): # ERROR HANDLING in case a song queries returns 'null' from musixmatch try: # GRAB the lyrics based on location (iloc) in chart_df song_search = chart_df.iloc[x,0] artist_search = chart_df.iloc[x,1] lyrics = musixmatch.matcher_lyrics_get(q_artist=artist_search, q_track=song_search)['message']['body']['lyrics']['lyrics_body'] # FORMATTING to truncate the nonsense at the end of the lyrics from MusixMatch song_length = len(lyrics) endpoint = len("******* This Lyrics is NOT for Commercial use *******\n(1409617829201)") lyrics = lyrics.replace("\n", " ") lyrics = str(lyrics[:song_length-endpoint]) # APPEND lyrics to lyrics_list lyrics_list.append(lyrics) except: error_counter = error_counter + 1 lyrics_list.append('MUSIXMATCH_NA') # CREATE new column in chart_df chart_df['Lyrics'] = lyrics_list # VIEW dataframe head chart_df.head() # REMOVE blanks and errors from the dataframe clean_chart_df = chart_df[(chart_df['Lyrics'] != "MUSIXMATCH_NA") & (chart_df['Lyrics'] != "")].reset_index(drop=True) # VIEW dataframe head clean_chart_df.head() # clean_chart_df.count() # Vader Sentiment Analysis conducted on each song in the dataframe # INITIALIZE a list to hold the sentiments lyrics_sentiments = [] # ANALYZE the list for y in range(len(clean_chart_df)): results = analyzer.polarity_scores(clean_chart_df.iloc[y,6]) compound = results["compound"] pos = results["pos"] neu = results["neu"] neg = results["neg"] lyrics_sentiments.append({"Compound": compound, "Positive": pos, "Negative": neg, "Neutral": neu}) # CREATE a dataframe of sentiment analysis that will be appended to the chart_df lyrics_sentiments_df = pd.DataFrame(lyrics_sentiments) # APPEND new columns containing the sentiment analysis clean_chart_df['Compound'] = lyrics_sentiments_df['Compound'] clean_chart_df['Positive'] = lyrics_sentiments_df['Positive'] clean_chart_df['Negative'] = lyrics_sentiments_df['Negative'] clean_chart_df['Neutral'] = lyrics_sentiments_df['Neutral'] # SAVE to a .csv output clean_chart_df.to_csv('billboard_analysis.csv') # VIEW resulting dataframe head clean_chart_df.head() # Create the pandas dataframe group in order to calculate averages by year chart_group = clean_chart_df.groupby(["Year"]).mean() chart_group_df = pd.DataFrame(chart_group).reset_index(drop=False) chart_group_df # TEST HISTOGRAM of negative sentiment results plt.figure(figsize=(20,20)) plt.subplot(2, 1, 2) plt.hist(chart_group_df['Negative'], 10, density=True, alpha=0.7, label="population1") x,labels=plt.xticks() labels=[0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, 0.1 , 0.11] plt.xticks(x,labels,fontsize=16) plt.xticks(label=labels) plt.yticks(fontsize=16) plt.title("Distribution of Songs by Negative Sentiment",fontsize=20) plt.savefig("histogram.png") # Create a local x-axis x_axis = np.arange(1968,2018) print(len(x_axis)) print(len(chart_group_df['Negative'])) # Scatterplot of negative sentiment analysis # Creates the regression line (slope, intercept, r_value, p_value, std_err) = linregress(x_axis, chart_group_df['Negative']) fit = slope * x_axis + intercept r2 = r_value ** 2 # Sets up plot fig, ax = plt.subplots(figsize=(20,10)) ax.set_xlabel("Years",fontsize=18) ax.set_ylabel("Sentiment Analysis: Negative",fontsize=18) ax.tick_params(labelsize=16) ax.set_title(label="Negative Sentiment over Time",fontsize=24) # Plots the data ax.plot(x_axis, chart_group_df['Negative'], marker='o', color=('red'), linewidth=0.5) ax.plot(x_axis, fit, 'b--') plt.savefig("negative.png") plt.show print(f'r = {r_value}') print(f'r^2 = {r2}') print(f'std err = {std_err}') print(f'p-value = {p_value}') print('49% of the variation in the dependent variable (negative sentiment score) is accounted for by the variation \ in the independent variable (time in years). We feel comfortable using R^2, because the data is approximately \ normally distributed based on the shape of the histogram (slightly skewed right). Additionally, the p-value \ is 0+ which is < 0.05. \ Conclusion: We reject the null hypothesis (Ho) in favor of the alternative. These two variables are strongly related') # Scatterplot of positive sentiment analysis # Creates the regression line (slope, intercept, r_value, p_value, std_err) = linregress(x_axis, chart_group_df['Positive']) fit = slope * x_axis + intercept r2 = r_value ** 2 # Sets up plot fig, ax = plt.subplots(figsize=(20,10)) ax.set_xlabel("Years",fontsize=18) ax.set_ylabel("Sentiment Analysis: Positive",fontsize=18) ax.tick_params(labelsize=16) ax.set_title(label="Positive Sentiment over Time",fontsize=24) # Plots the data ax.plot(x_axis, chart_group_df['Positive'], marker='o', color=('green'), linewidth=0.5) ax.plot(x_axis, fit, 'b--') plt.savefig("positive.png") plt.show print(f'r = {r_value}') print(f'r^2 = {r2}') print(f'std err = {std_err}') print(f'p-value = {p_value}') # Scatterplot of neutral sentiment analysis # Creates the regression line (slope, intercept, r_value, p_value, std_err) = linregress(x_axis, chart_group_df['Neutral']) fit = slope * x_axis + intercept r2 = r_value ** 2 # Sets up plot fig, ax = plt.subplots(figsize=(20,10)) ax.set_xlabel("Years",fontsize=18) ax.set_ylabel("Sentiment Analysis: Neutral",fontsize=18) ax.tick_params(labelsize=16) ax.set_title(label="Neutral Sentiment over Time",fontsize=24) # Plots the data ax.plot(x_axis, chart_group_df['Neutral'], marker='o', color=('blue'), linewidth=0.5) ax.plot(x_axis, fit, 'b--') plt.savefig("neutral.png") plt.show print(f'r = {r_value}') print(f'r^2 = {r2}') print(f'std err = {std_err}') print(f'p-value = {p_value}') # Scatterplot of compound sentiment analysis # Creates the regression line (slope, intercept, r_value, p_value, std_err) = linregress(x_axis, chart_group_df['Compound']) fit = slope * x_axis + intercept r2 = r_value ** 2 # Sets up plot fig, ax = plt.subplots(figsize=(20,10)) ax.set_xlabel("Years",fontsize=18) ax.set_ylabel("Sentiment Analysis: Compound",fontsize=18) ax.tick_params(labelsize=16) ax.set_title(label="Compound Sentiment over Time",fontsize=24) # Plots the data ax.plot(x_axis, chart_group_df['Compound'], marker='x', color=('black'), linewidth=0.5) ax.plot(x_axis, fit, 'b--') plt.savefig("compound.png") plt.show print(f'r = {r_value}') print(f'r^2 = {r2}') print(f'std err = {std_err}') print(f'p-value = {p_value}')<jupyter_output><empty_output>
no_license
/projects/Lyrics-Analyzer/Output/Test Samples/JB_Sample/.ipynb_checkpoints/Project_Notebook_FINAL_JB-checkpoint.ipynb
jeffreybox/GTATL201805DATA3
1
<jupyter_start><jupyter_text>### Test-train split<jupyter_code>feature_columns = 'average_stars review_count compliment_cool compliment_cute \ compliment_funny compliment_hot compliment_list compliment_more \ compliment_note compliment_photos compliment_plain cool fans \ funny rev_length rev_stars rev_use friend_count friend_label \ rev_count_label buss_star buss_review alpha_length length_lemmatize \ NN NNP NNS PDT POS PRP PRP$ RB RBR RBS RP VB \ VBD VBG VBN VBP VBZ WDT WP WRB'.split() X = (cleaned_data[feature_columns]) X.replace(to_replace= 'NaN', value=np.nan, inplace=True) X.fillna(value= 0, inplace=True) y= cleaned_data['Faker'] X_train=X.values[:160,:] X_test=X.values[160:,] y_train =y.values[:160,] y_test =y.values[160:,] <jupyter_output>/home/titli/anaconda3/lib/python3.6/site-packages/pandas/core/frame.py:3795: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy method=method) /home/titli/anaconda3/lib/python3.6/site-packages/pandas/core/frame.py:3787: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy downcast=downcast, **kwargs) <jupyter_text>### Feature selection<jupyter_code>from sklearn.ensemble import RandomForestClassifier RFcw = RandomForestClassifier(random_state=40, n_estimators = 50, min_samples_leaf=5, class_weight="balanced") RFcw.fit(X_train, y_train) # Feature importance #forest = RFcw.best_estimator_.steps[-1][1] imps = RFcw.feature_importances_ # Std dev due to all trees in the forest std = np.std([tree.feature_importances_ for tree in RFcw.estimators_], axis=0) # Index that sorts the features with most important first indices = np.argsort(imps)[::-1] for f in range(len(feature_columns)): print(f+1, feature_columns[indices[f]], imps[indices[f]]) selected_feature = [] for i in range(20): selected_feature.append(feature_columns[indices[i]]) <jupyter_output>1 PRP$ 0.09078578784594421 2 average_stars 0.0775395134909361 3 NN 0.06063926493135878 4 buss_review 0.05987820737856189 5 VB 0.05971976446595563 6 review_count 0.05706733732029779 7 rev_length 0.05095355081116822 8 VBD 0.05039938581656459 9 VBZ 0.04727061028771006 10 RB 0.039802843884653445 11 length_lemmatize 0.03722743532116322 12 PRP 0.037204700636598254 13 rev_stars 0.03366753528498414 14 NNS 0.03223270609026408 15 alpha_length 0.03171755626693837 16 VBG 0.03158601665967478 17 WRB 0.031040453129827304 18 VBN 0.01970257065924371 19 buss_star 0.019024365506568573 20 friend_count 0.016763321124830893 21 RP 0.01572484918826479 22 rev_use 0.015040555502168356 23 VBP 0.01351162025111022 24 rev_count_label 0.009282615290685692 25 compliment_funny 0.009172669308572233 26 cool 0.008406717354263342 27 friend_label 0.007913955599181727 28 WDT 0.0070361235106931016 29 POS 0.005554062762930463 30 funny 0.005427997432628426 31 compliment_plain 0.0051176937910191135 32 fans 0.0034094091042426644[...]<jupyter_text>### New test-train split based on selected features<jupyter_code>X_new = (cleaned_data[selected_feature]) X_new_train=X_new.values[:160,:] X_new_test=X_new.values[160:,] from sklearn.naive_bayes import GaussianNB from sklearn.naive_bayes import MultinomialNB from imblearn.pipeline import Pipeline as pl from imblearn.under_sampling import RandomUnderSampler steps = [("rus", RandomUnderSampler(random_state=30)), ("model", MultinomialNB(alpha=100))] GNBrus1 = pl(steps) GNBrus1.fit(X_new_train, y_train) y_pred =GNBrus1.predict(X_new_test) y_pred_train =GNBrus1.predict(X_new_train) def confusion(predictions, actuals): actuals=actuals.values[:,0] if isinstance(actuals,pd.DataFrame) else actuals true_pos= (predictions==1) & (actuals==1) true_pos.sum() true_neg= (predictions==0) & (actuals==0) true_neg.sum() false_pos= (predictions==1) & (actuals==0) false_pos.sum() false_neg= (predictions==0) & (actuals==1) false_neg.sum() prec=true_pos.sum()/(true_pos.sum()+false_pos.sum()) accur=(true_pos.sum()+true_neg.sum())/(true_pos.sum()+false_pos.sum()+ \ true_neg.sum()+ false_neg.sum()) recall = true_pos.sum()/(true_pos.sum()+false_neg.sum()) F1=2*(prec*recall/(prec+recall)) return(true_pos.sum(), false_pos.sum(),false_neg.sum(),true_neg.sum(), accur,recall, prec, F1) np.array((confusion(y_pred_train, y_train.reshape((1, 160))))) confusion(y_pred, y_test.reshape((1, 40))) from sklearn.ensemble import ExtraTreesClassifier from sklearn.preprocessing import MinMaxScaler from sklearn.model_selection import GridSearchCV from imblearn.pipeline import Pipeline as pl from imblearn.under_sampling import RandomUnderSampler from sklearn.metrics import average_precision_score PRAUC = [] for trees in [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]: ETcwiter = ExtraTreesClassifier(random_state=40, n_estimators = trees, min_samples_leaf=4, class_weight="balanced") ETcwiter.fit(X_new_train, y_train) PRAUC.append(average_precision_score(y_test, ETcwiter.predict_proba(X_new_test)[:,1])) plt.scatter(x= [10, 20, 30, 40, 50, 60, 70, 80, 90, 100], y= PRAUC, marker='^', color='red') plt.xlabel('No. of trees', fontsize=15) plt.ylabel('average_precision_score',fontsize=15) plt.title('Best estimator for no. of trees in extra tress',fontsize=15) ETcwiter = ExtraTreesClassifier(random_state=40, n_estimators = 100, min_samples_leaf=4, class_weight="balanced") ETcwiter.fit(X_new_train, y_train) y_pred = ETcwiter.predict(X_new_test) y_pred_train = ETcwiter.predict(X_new_train) np.array((confusion(y_pred, y_test.reshape((1, 40))))) np.array((confusion(y_pred_train, y_train.reshape((1, 160)))))<jupyter_output><empty_output>
no_license
/Capstone_project_fake_reviewers/Machine_learning/Naivebayes_extratree.ipynb
tannisthamaiti/Fake-reviewers-in-Yelp
3
<jupyter_start><jupyter_text>--- title: "Point-in-time joins with PySpark" date: 2021-09-09 type: technical_note draft: false ---# Point-in-Time (PIT) joins in Hopsworks Feature Store In order to create a training dataset, data scientists usually have to generate information about the future by putting themselves back in time, or travelling back in time. Let's take the case of churn prediction: We want to predict which of our users will churn within the next few months. To train such a model we need to construct a training dataset containing rows for each of our users, one column indicating whether a user churned and X additional columns with features about the user, such as his historical purchases or interactions with the product. Since we don't know the future yet (it's what we want to predict), to generate such prediction targets, we have to go back in time and determine which users churned in the last couple of months. ### Snapshot based time travel If you have looked at our other time travel notebooks, the most simple solution to this problem would be to choose a single cutoff time for all our users. Using the time travel capabilities of the Feature Store with Apache Hudi make it easy enough to fetch data from a single point in time. This can work if features aren't frequently updated, however, when prediction target events of different users lie far apart and features are updated frequently you might leave a lot of information unused, hurting the accuracy of your model. ### Individual point-in-time correct cutoffs A better approach includes using all information up to the time the prediction target event happened. So that means we still go back, let's say a maximum 6 months, in time, however, we want to remember the prediction target event time stamp on a user level (row level) and find the latest values of our prediction features before this point in time. The following diagrams illustrates this approach for one user: ![image.png](attachment:9ada7a9e-021e-47ed-9993-db2c6a6d8c32.png) A problem often occurring at this stage is the possibility of leaking information from the future (light red signals) into the training dataset. This means using feature signals which happened after the prediction target event has to be strictly avoided. What we want to retrieve from the feature store are the green, and the green only, feature signals. In this case it is not as simple as performing a time travel query with a single time stamp. This challenge is solved by a point-in-time correct join instead. **Point-in-time joins** prevent feature leakage by recreating the state of the world at a single point in time for every entity or primary key value (user in our case). **Hopsworks Feature Store** abstracts this complexity away by simply telling it where to find the relevant event time stamps for feature groups. We will go through the process in the rest of the notebook. ## Event-time enabled Feature GroupsFor the Feature Store to be able to perform a PIT join, we need to tell it where to find the event time stamp within each feature group. Event-time is a timestamp indicating the instant in time when an event happened at the source of the event, so this is *not* an ingestion time stamp or the like, but instead should originate from your source systems. To "event-time enable" a feature group, you set the `event_time` argument at feature group creation. We are using simple Integers to indicate the timestamps, for better readability. ### For simplicity we will create three feature groups * `marketing` with customer id, complaints in the last 7 days, outbound activities in the last 7 days and coupons received in the last 14 days; * `contract` with customer id, contract_type; * `churn` which will contain labels wether loan was cancelled `1` or rejected `0` with a timestamp when the contract was created or cancelled, note that this feature group has a composite primary key of customer id and a contract id, referring to the id in the contract feature group.<jupyter_code>import hsfs from pyspark.sql import DataFrame, Row from pyspark.sql.types import * connection = hsfs.connection() # get a reference to the feature store, you can access also shared feature stores by providing the feature store name fs = connection.get_feature_store() marketing_schema = StructType([ StructField("customer_id", IntegerType(), True), StructField("ts", LongType(), True), StructField("complaints_7d", IntegerType(), True), StructField("outbound_7d", IntegerType(), True), StructField("coupon_14d", IntegerType(), True) ]) contract_schema = StructType([ StructField("contract_id", IntegerType(), True), StructField("ts", LongType(), True), StructField("contract_type", StringType(), True) ]) churn_schema = StructType([ StructField("customer_id", IntegerType(), True), StructField("contract_id", IntegerType(), True), StructField("ts", LongType(), True), StructField("contract_churn", IntegerType(), True) ])<jupyter_output><empty_output><jupyter_text>We will first load the Churn Feature Group with the initial contracts. We can assume there is a job running inserting new rows for every new contract into this feature group.<jupyter_code>new_contracts = [ Row(1, 100, 10010, 0), Row(2, 101, 10017, 0), Row(3, 102, 10035, 0), Row(4, 103, 10023, 0), Row(5, 104, 10546, 0), Row(6, 105, 10213, 0), Row(7, 106, 10056, 0), Row(8, 107, 10012, 0) ] new_contracts_df = spark.createDataFrame(new_contracts, churn_schema)<jupyter_output><empty_output><jupyter_text>At the same time some contracts will be cancelled and inserted into the feature group over time. We will perform this insertion with a bit of time difference, so that we can later demonstrate the capabilities of PIT joins together with time-travel queries.<jupyter_code>churned_contracts = [ Row(1, 100, 10356, 1), Row(5, 104, 10692, 1), Row(6, 105, 10375, 1), Row(8, 107, 10023, 1) ] churned_contracts_df = spark.createDataFrame(churned_contracts, churn_schema)<jupyter_output><empty_output><jupyter_text>Now let's create some mock data for the secondary feature groups, containing the actual explanatory features used to predict the churn. The contract feature group is a feature group that gets new contracts appended with information about the contract itself only, such as the type of contract. Hence all timestamps are the same as the inital contract creation in the churn feature group.<jupyter_code>contracts = [ Row(100, 10010, "Long-term"), Row(101, 10017, "Short-term"), Row(102, 10035, "Trial"), Row(103, 10023, "Short-term"), Row(104, 10546, "Long-term"), Row(105, 10213, "Trial"), Row(106, 10056, "Long-term"), Row(107, 10012, "Short-term") ] contracts_df = spark.createDataFrame(contracts, contract_schema)<jupyter_output><empty_output><jupyter_text>The marketing activities feature group contains features related to outbound and inbound contacts to the customer. You can imagine these to be computed by a streaming application, updating the features every time new events arrive. In the point in time join we want to get the latest of these updates just before or at the same time as the prediction event happened. Contracts can be in the training dataset twice, once when they haven't churned yet and then when they churned. The rows which should be picked for either of the target events are marked with a comment. At the same time, always the latest state of the customer's marketing profile should be available in the online feature store.<jupyter_code>marketing = [ Row(1, 10010, 0, 0, 1), # this one Row(1, 10174, 3, 0, 4), Row(1, 10257, 7, 0, 3), Row(1, 10352, 3, 0, 5), # this one Row(1, 10753, 0, 0, 0), Row(1, 10826, 0, 0, 1), # online feature store Row(2, 10017, 0, 1, 1), # this one Row(2, 10021, 0, 1, 1), Row(2, 10034, 0, 1, 2), # online feature store Row(3, 10035, 1, 3, 0), # this one Row(3, 10275, 1, 2, 0), Row(5, 10546, 0, 1, 0), # this one Row(5, 10598, 2, 2, 1), # this one Row(5, 13567, 0, 1, 0), Row(5, 16245, 0, 1, 0), # online feature store Row(6, 10213, 0, 0, 1), # this one Row(6, 10234, 0, 5, 0), # this one Row(6, 10436, 0, 0, 1), # online feature store Row(7, 10056, 0, 0, 0), # this one Row(7, 10056, 0, 1, 0), Row(7, 10056, 0, 2, 1), Row(7, 10056, 0, 3, 0), # online feature store Row(8, 10012, 0, 0, 1), # this one Row(8, 10023, 10, 0, 1), # this one Row(8, 10033, 0, 0, 1), # online feature store ] marketing_df = spark.createDataFrame(marketing, marketing_schema)<jupyter_output><empty_output><jupyter_text>### Create the feature groupsWe are now ready to create our three feature groups. Note the additional argument we are passing, in order to tell the Feature Store which column should be used as `event_time`.<jupyter_code>marketing_fg = fs.create_feature_group("marketing", version=1, description="Features about inbound/outbound communication with customers", online_enabled=True, statistics_config=False, primary_key=["customer_id"], event_time="ts") marketing_fg.save(marketing_df) contracts_fg = fs.create_feature_group("contracts", version=1, description="Contract information features", online_enabled=True, statistics_config=False, primary_key=["contract_id"], event_time="ts") contracts_fg.save(contracts_df) churn_fg = fs.create_feature_group("churn", version=1, description="Customer/contract information about activity of contract", online_enabled=True, statistics_config=False, primary_key=["customer_id", "contract_id"], event_time="ts") churn_fg.save(new_contracts_df)<jupyter_output><empty_output><jupyter_text>We insert the churned contracts, in a separate upsert step.<jupyter_code>churn_fg.insert(churned_contracts_df)<jupyter_output>StorageWarning: The statistics are not enabled of feature group `churn`, with version `1`. No statistics computed.<jupyter_text>### Constructing a Point-in-Time Join The Feature Store HSFS API comes with a `Query` abstraction. Operations such as `.join` or `.select` on a feature group return such `Query` objects. Queries are solely based on metadata up until they are saved as training datasets or read into Dataframes. There are **two requirements** to construct a point in time join: 1. All feature groups have to be event time enabled. If there is no event timestamp available, you can fall back on creating an ingestion timestamp in your feature engineering pipeline. 2. The label (feature) group is the Feature Group that contains the column that is defined as the "label" column when creating the training dataset. The **left-most feature group has to be the label group**, meaning this is the feature group of which the timestamps will be used as reference timestamp against which the explaining features are joined (dark red dots in the diagram above). In our case this is the churn feature group. So let's construct the join:<jupyter_code>query = churn_fg.select_all().join(contracts_fg.select(["contract_type"]), on=["contract_id"]) \ .join(marketing_fg.select(["complaints_7d", "outbound_7d", "coupon_14d"]), on=["customer_id"])<jupyter_output><empty_output><jupyter_text>You can print and look at the constructed SQL join query. However, there is no need to go into this complexity, as it will all be handled by the feature store.<jupyter_code>print(query.to_string())<jupyter_output>WITH right_fg0 AS (SELECT * FROM (SELECT `fg2`.`customer_id`, `fg2`.`contract_id`, `fg2`.`ts`, `fg2`.`contract_churn`, `fg0`.`contract_type`, RANK() OVER (PARTITION BY `fg2`.`contract_id`, `fg2`.`ts` ORDER BY `fg0`.`ts` DESC) pit_rank_hopsworks FROM `fg2` `fg2` INNER JOIN `fg0` `fg0` ON `fg2`.`contract_id` = `fg0`.`contract_id` AND `fg2`.`ts` >= `fg0`.`ts`) NA WHERE `pit_rank_hopsworks` = 1), right_fg1 AS (SELECT * FROM (SELECT `fg2`.`customer_id`, `fg2`.`contract_id`, `fg2`.`ts`, `fg2`.`contract_churn`, `fg1`.`complaints_7d`, `fg1`.`outbound_7d`, `fg1`.`coupon_14d`, RANK() OVER (PARTITION BY `fg2`.`customer_id`, `fg2`.`ts` ORDER BY `fg1`.`ts` DESC) pit_rank_hopsworks FROM `fg2` `fg2` INNER JOIN `fg1` `fg1` ON `fg2`.`customer_id` = `fg1`.`customer_id` AND `fg2`.`ts` >= `fg1`.`ts`) NA WHERE `pit_rank_hopsworks` = 1) (SELECT `right_fg0`.`customer_id`, `right_fg0`.`contract_id`, `right_fg0`.`ts`, `right_fg0`.`contract_churn`, `right_fg0`.`contract_type`, `right_fg1`.`complaints_7d`, `righ[...]<jupyter_text>As explained above, the Query itself does not perform any join yet until you call an action such as `.read` on it:<jupyter_code>query.read().show(20)<jupyter_output>+-----------+-----------+-----+--------------+-------------+-------------+-----------+----------+ |customer_id|contract_id| ts|contract_churn|contract_type|complaints_7d|outbound_7d|coupon_14d| +-----------+-----------+-----+--------------+-------------+-------------+-----------+----------+ | 6| 105|10375| 1| Trial| 0| 5| 0| | 5| 104|10692| 1| Long-term| 2| 2| 1| | 5| 104|10546| 0| Long-term| 0| 1| 0| | 2| 101|10017| 0| Short-term| 0| 1| 1| | 7| 106|10056| 0| Long-term| 0| 0| 0| | 8| 107|10023| 1| Short-term| 10| 0| 1| | 1| 100|10356| 1| Long-term| 3| 0| 5| | 1| [...]<jupyter_text>#### Filters on point-in-time queries As with any other query, it is possible to apply filters to it. For example, if you are planning to train only a model for a certain contract type:<jupyter_code>churn_fg.select_all().join( contracts_fg.select(["contract_type"]) .filter(contracts_fg.contract_type == "Long-term") , on=["contract_id"]) \ .join(marketing_fg.select(["complaints_7d", "outbound_7d", "coupon_14d"]), on=["customer_id"]).read().show(20)<jupyter_output>+-----------+-----------+-----+--------------+-------------+-------------+-----------+----------+ |customer_id|contract_id| ts|contract_churn|contract_type|complaints_7d|outbound_7d|coupon_14d| +-----------+-----------+-----+--------------+-------------+-------------+-----------+----------+ | 5| 104|10692| 1| Long-term| 2| 2| 1| | 5| 104|10546| 0| Long-term| 0| 1| 0| | 7| 106|10056| 0| Long-term| 0| 0| 0| | 1| 100|10356| 1| Long-term| 3| 0| 5| | 1| 100|10010| 0| Long-term| 0| 0| 1| +-----------+-----------+-----+--------------+-------------+-------------+-----------+----------+<jupyter_text>#### Combining time-travel and point-in-time join in one query We performed a separate upsert on our churn feature group in order to be able to demonstrate time-travel, so let's look at the commit information:<jupyter_code>commit_details = churn_fg.commit_details() print(commit_details)<jupyter_output>{1631088747000: {'committedOn': '20210908081227', 'rowsUpdated': 0, 'rowsInserted': 4, 'rowsDeleted': 0}, 1631088722000: {'committedOn': '20210908081202', 'rowsUpdated': 0, 'rowsInserted': 8, 'rowsDeleted': 0}}<jupyter_text>As you can see, there are two commits: (1) the initial commit, which inserted the eight contracts. And (2) the upsert commit with the additinal four churned contracts. Now we would like to query the state of the Feature Store at the time before the churned contracts were ingested:<jupyter_code># getting the correct timestamp, we sort by the commit time and get the string representation of the commit time committed_on = commit_details[sorted(commit_details)[0]]["committedOn"] committed_on contracts_fg.commit_details() churn_fg.select_all().as_of(committed_on).read().show(20)<jupyter_output>+-----------+-----------+-----+--------------+ |customer_id|contract_id| ts|contract_churn| +-----------+-----------+-----+--------------+ | 2| 101|10017| 0| | 5| 104|10546| 0| | 8| 107|10012| 0| | 3| 102|10035| 0| | 6| 105|10213| 0| | 7| 106|10056| 0| | 4| 103|10023| 0| | 1| 100|10010| 0| +-----------+-----------+-----+--------------+<jupyter_text>And combining this with our point-in-time join:<jupyter_code>query.as_of(committed_on).read().show(20)<jupyter_output>+-----------+-----------+-----+--------------+-------------+-------------+-----------+----------+ |customer_id|contract_id| ts|contract_churn|contract_type|complaints_7d|outbound_7d|coupon_14d| +-----------+-----------+-----+--------------+-------------+-------------+-----------+----------+ | 5| 104|10546| 0| Long-term| 0| 1| 0| | 2| 101|10017| 0| Short-term| 0| 1| 1| | 7| 106|10056| 0| Long-term| 0| 0| 0| | 1| 100|10010| 0| Long-term| 0| 0| 1| | 6| 105|10213| 0| Trial| 0| 0| 1| | 8| 107|10012| 0| Short-term| 0| 0| 1| | 3| 102|10035| 0| Trial| 1| 3| 0| +-----------+-------[...]<jupyter_text>### Creating the training dataset Creating the training dataset is now as simple as initializing the metadata with `.create_training_dataset()` and subsequently persisting it and materializing the query with `.save()`.<jupyter_code>td = fs.create_training_dataset("churn_model", version=1, data_format="csv", label=["contract_churn"]) td.save(query)<jupyter_output><empty_output><jupyter_text>#### Querying the online feature store We can reproduce the query and get the latest feature vector for each contract from the online feature store now. **NOTE:** - Any applied filter at the time of creating the training dataset is not applied during serving time. It is the responsibility of the application performing the lookup, to only provide contract ids which belong for example to a certain category. The reason for that is that we always want to get back a vector, applying a filter, might lead to no results.<jupyter_code>td.get_serving_vector({"customer_id": 1, "contract_id": 100}) td.get_serving_vector({"customer_id": 2, "contract_id": 101})<jupyter_output>[2, 101, 10017, 'Short-term', 0, 1, 2]
no_license
/notebooks/featurestore/hsfs/time_travel/point_in_time_join_python.ipynb
robzor92/hops-examples
16
<jupyter_start><jupyter_text># 範例 : 計程車費率預測 https://www.kaggle.com/c/new-york-city-taxi-fare-prediction# [作業目標] - 試著模仿範例寫法, 使用程車費率預測競賽練習時間欄位處理# [作業重點] - 新增星期幾(day of week)與第幾周(week of year)這兩項特徵, 觀察有什麼影響 (In[4], Out[4], In[5], Out[5]) - 新增加上年週期與周周期特徵(參考教材) , 觀察有什麼影響 (In[8], Out[8], In[9], Out[9]) <jupyter_code># 做完特徵工程前的所有準備 import pandas as pd import numpy as np import datetime from sklearn.preprocessing import MinMaxScaler from sklearn.model_selection import cross_val_score from sklearn.linear_model import LinearRegression from sklearn.ensemble import GradientBoostingRegressor data_path = 'data/' df = pd.read_csv(data_path + 'taxi_data1.csv') train_Y = df['fare_amount'] df = df.drop(['fare_amount'] , axis=1) df.head() # 時間特徵分解方式:使用datetime df['pickup_datetime'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strptime(x, '%Y-%m-%d %H:%M:%S UTC')) df['pickup_year'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%Y')).astype('int64') df['pickup_month'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%m')).astype('int64') df['pickup_day'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%d')).astype('int64') df['pickup_hour'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%H')).astype('int64') df['pickup_minute'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%M')).astype('int64') df['pickup_second'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%S')).astype('int64') df.head() # 將結果使用線性迴歸 / 梯度提升樹分別看結果 df_temp = df.drop(['pickup_datetime'] , axis=1) scaler = MinMaxScaler() train_X = scaler.fit_transform(df_temp) Linear = LinearRegression() print(f'Linear Reg Score : {cross_val_score(Linear, train_X, train_Y, cv=5).mean()}') GDBT = GradientBoostingRegressor() print(f'Gradient Boosting Reg Score : {cross_val_score(GDBT, train_X, train_Y, cv=5).mean()}')<jupyter_output>Linear Reg Score : 0.026876871475640173 Gradient Boosting Reg Score : 0.7115447416761242 <jupyter_text># 作業1 * 對照範例,試著加入星期幾 (day of week) 與第幾周 (week of year) 這兩項特徵, 看看結果會比原本只有時間特徵分解的結果更好或更差? > 加入星期幾與第幾周的特徵後, 看起來結果似乎更差了, 可能表示"第幾周"這個特徵沒有提供多少資訊 但是搭配日週期之後效果有提升, 可能是假日與非假日(星期幾)的作息(日週期)與費率有相關性<jupyter_code># 語法參閱 https://docs.python.org/3/library/datetime.html df['pickup_dow'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%w')).astype('int64') df['pickup_woy'] = df['pickup_datetime'].apply(lambda x: datetime.datetime.strftime(x, '%W')).astype('int64') df.head() # 將結果使用線性迴歸 / 梯度提升樹分別看結果 df_temp = df.drop(['pickup_datetime'] , axis=1) train_X = scaler.fit_transform(df_temp) print(f'Linear Reg Score : {cross_val_score(Linear, train_X, train_Y, cv=5).mean()}') print(f'Gradient Boosting Reg Score : {cross_val_score(GDBT, train_X, train_Y, cv=5).mean()}') # 加上"日週期"特徵 (參考講義"週期循環特徵") import math df['day_cycle'] = df['pickup_hour']/12 + df['pickup_minute']/720 + df['pickup_second']/43200 df['day_cycle'] = df['day_cycle'].map(lambda x:math.sin(x*math.pi)) df.head() # 將結果使用線性迴歸 / 梯度提升樹分別看結果 df_temp = df.drop(['pickup_datetime'] , axis=1) train_X = scaler.fit_transform(df_temp) print(f'Linear Reg Score : {cross_val_score(Linear, train_X, train_Y, cv=5).mean()}') print(f'Gradient Boosting Reg Score : {cross_val_score(GDBT, train_X, train_Y, cv=5).mean()}')<jupyter_output>Linear Reg Score : 0.02827400966881246 Gradient Boosting Reg Score : 0.7126037331189518 <jupyter_text># 作業2 * 對照範例的日週期效果,試著參考投影片完成年週期與周週期的特徵 (也可以用你自己想到的方式), 看看結果會比範例中的結果更好或更差? > 從結果來看, 加上年週期與周週期, 正確性有顯著的提升, 這可能表示冷熱等確實與費率也有相關性(天氣熱比較搭計程車?)<jupyter_code># 加上"年週期"與"周週期"特徵 df['year_cycle'] = df['pickup_month']/6 + df['pickup_day']/180 df['year_cycle'] = df['year_cycle'].map(lambda x:math.cos(x*math.pi)) df['week_cycle'] = df['pickup_dow']/3.5 + df['pickup_hour']/84 df['week_cycle'] = df['week_cycle'].map(lambda x:math.sin(x*math.pi)) df.head() # 將結果使用線性迴歸 / 梯度提升樹分別看結果 df_temp = df.drop(['pickup_datetime'] , axis=1) train_X = scaler.fit_transform(df_temp) print(f'Linear Reg Score : {cross_val_score(Linear, train_X, train_Y, cv=5).mean()}') print(f'Gradient Boosting Reg Score : {cross_val_score(GDBT, train_X, train_Y, cv=5).mean()}')<jupyter_output>Linear Reg Score : 0.028326599866693947 Gradient Boosting Reg Score : 0.7154288505120123
no_license
/answers/Day_025_DayTime_Features_Ans.ipynb
Julian-Chu/2nd-ML100Days
3