Datasets:
Delete loading script
Browse files
mlqa.py
DELETED
@@ -1,206 +0,0 @@
|
|
1 |
-
"""TODO(mlqa): Add a description here."""
|
2 |
-
|
3 |
-
|
4 |
-
import json
|
5 |
-
import os
|
6 |
-
|
7 |
-
import datasets
|
8 |
-
|
9 |
-
|
10 |
-
# TODO(mlqa): BibTeX citation
|
11 |
-
_CITATION = """\
|
12 |
-
@article{lewis2019mlqa,
|
13 |
-
title={MLQA: Evaluating Cross-lingual Extractive Question Answering},
|
14 |
-
author={Lewis, Patrick and Oguz, Barlas and Rinott, Ruty and Riedel, Sebastian and Schwenk, Holger},
|
15 |
-
journal={arXiv preprint arXiv:1910.07475},
|
16 |
-
year={2019}
|
17 |
-
}
|
18 |
-
"""
|
19 |
-
|
20 |
-
# TODO(mlqa):
|
21 |
-
_DESCRIPTION = """\
|
22 |
-
MLQA (MultiLingual Question Answering) is a benchmark dataset for evaluating cross-lingual question answering performance.
|
23 |
-
MLQA consists of over 5K extractive QA instances (12K in English) in SQuAD format in seven languages - English, Arabic,
|
24 |
-
German, Spanish, Hindi, Vietnamese and Simplified Chinese. MLQA is highly parallel, with QA instances parallel between
|
25 |
-
4 different languages on average.
|
26 |
-
"""
|
27 |
-
_URL = "https://dl.fbaipublicfiles.com/MLQA/"
|
28 |
-
_DEV_TEST_URL = "MLQA_V1.zip"
|
29 |
-
_TRANSLATE_TEST_URL = "mlqa-translate-test.tar.gz"
|
30 |
-
_TRANSLATE_TRAIN_URL = "mlqa-translate-train.tar.gz"
|
31 |
-
_LANG = ["ar", "de", "vi", "zh", "en", "es", "hi"]
|
32 |
-
_TRANSLATE_LANG = ["ar", "de", "vi", "zh", "es", "hi"]
|
33 |
-
|
34 |
-
|
35 |
-
class MlqaConfig(datasets.BuilderConfig):
|
36 |
-
def __init__(self, data_url, **kwargs):
|
37 |
-
"""BuilderConfig for MLQA
|
38 |
-
|
39 |
-
Args:
|
40 |
-
data_url: `string`, url to the dataset
|
41 |
-
**kwargs: keyword arguments forwarded to super.
|
42 |
-
"""
|
43 |
-
super(MlqaConfig, self).__init__(
|
44 |
-
version=datasets.Version(
|
45 |
-
"1.0.0",
|
46 |
-
),
|
47 |
-
**kwargs,
|
48 |
-
)
|
49 |
-
self.data_url = data_url
|
50 |
-
|
51 |
-
|
52 |
-
class Mlqa(datasets.GeneratorBasedBuilder):
|
53 |
-
"""TODO(mlqa): Short description of my dataset."""
|
54 |
-
|
55 |
-
# TODO(mlqa): Set up version.
|
56 |
-
VERSION = datasets.Version("1.0.0")
|
57 |
-
BUILDER_CONFIGS = (
|
58 |
-
[
|
59 |
-
MlqaConfig(
|
60 |
-
name="mlqa-translate-train." + lang,
|
61 |
-
data_url=_URL + _TRANSLATE_TRAIN_URL,
|
62 |
-
description="Machine-translated data for Translate-train (SQuAD Train and Dev sets machine-translated into "
|
63 |
-
"Arabic, German, Hindi, Vietnamese, Simplified Chinese and Spanish)",
|
64 |
-
)
|
65 |
-
for lang in _LANG
|
66 |
-
if lang != "en"
|
67 |
-
]
|
68 |
-
+ [
|
69 |
-
MlqaConfig(
|
70 |
-
name="mlqa-translate-test." + lang,
|
71 |
-
data_url=_URL + _TRANSLATE_TEST_URL,
|
72 |
-
description="Machine-translated data for Translate-Test (MLQA-test set machine-translated into English) ",
|
73 |
-
)
|
74 |
-
for lang in _LANG
|
75 |
-
if lang != "en"
|
76 |
-
]
|
77 |
-
+ [
|
78 |
-
MlqaConfig(
|
79 |
-
name="mlqa." + lang1 + "." + lang2,
|
80 |
-
data_url=_URL + _DEV_TEST_URL,
|
81 |
-
description="development and test splits",
|
82 |
-
)
|
83 |
-
for lang1 in _LANG
|
84 |
-
for lang2 in _LANG
|
85 |
-
]
|
86 |
-
)
|
87 |
-
|
88 |
-
def _info(self):
|
89 |
-
# TODO(mlqa): Specifies the datasets.DatasetInfo object
|
90 |
-
return datasets.DatasetInfo(
|
91 |
-
# This is the description that will appear on the datasets page.
|
92 |
-
description=_DESCRIPTION,
|
93 |
-
# datasets.features.FeatureConnectors
|
94 |
-
features=datasets.Features(
|
95 |
-
{
|
96 |
-
"context": datasets.Value("string"),
|
97 |
-
"question": datasets.Value("string"),
|
98 |
-
"answers": datasets.features.Sequence(
|
99 |
-
{"answer_start": datasets.Value("int32"), "text": datasets.Value("string")}
|
100 |
-
),
|
101 |
-
"id": datasets.Value("string"),
|
102 |
-
# These are the features of your dataset like images, labels ...
|
103 |
-
}
|
104 |
-
),
|
105 |
-
# If there's a common (input, target) tuple from the features,
|
106 |
-
# specify them here. They'll be used if as_supervised=True in
|
107 |
-
# builder.as_dataset.
|
108 |
-
supervised_keys=None,
|
109 |
-
# Homepage of the dataset for documentation
|
110 |
-
homepage="https://github.com/facebookresearch/MLQA",
|
111 |
-
citation=_CITATION,
|
112 |
-
)
|
113 |
-
|
114 |
-
def _split_generators(self, dl_manager):
|
115 |
-
"""Returns SplitGenerators."""
|
116 |
-
# TODO(mlqa): Downloads the data and defines the splits
|
117 |
-
# dl_manager is a datasets.download.DownloadManager that can be used to
|
118 |
-
# download and extract URLs
|
119 |
-
if self.config.name.startswith("mlqa-translate-train"):
|
120 |
-
archive = dl_manager.download(self.config.data_url)
|
121 |
-
lang = self.config.name.split(".")[-1]
|
122 |
-
return [
|
123 |
-
datasets.SplitGenerator(
|
124 |
-
name=datasets.Split.TRAIN,
|
125 |
-
# These kwargs will be passed to _generate_examples
|
126 |
-
gen_kwargs={
|
127 |
-
"filepath": f"mlqa-translate-train/{lang}_squad-translate-train-train-v1.1.json",
|
128 |
-
"files": dl_manager.iter_archive(archive),
|
129 |
-
},
|
130 |
-
),
|
131 |
-
datasets.SplitGenerator(
|
132 |
-
name=datasets.Split.VALIDATION,
|
133 |
-
# These kwargs will be passed to _generate_examples
|
134 |
-
gen_kwargs={
|
135 |
-
"filepath": f"mlqa-translate-train/{lang}_squad-translate-train-dev-v1.1.json",
|
136 |
-
"files": dl_manager.iter_archive(archive),
|
137 |
-
},
|
138 |
-
),
|
139 |
-
]
|
140 |
-
|
141 |
-
else:
|
142 |
-
if self.config.name.startswith("mlqa."):
|
143 |
-
dl_file = dl_manager.download_and_extract(self.config.data_url)
|
144 |
-
name = self.config.name.split(".")
|
145 |
-
l1, l2 = name[1:]
|
146 |
-
return [
|
147 |
-
datasets.SplitGenerator(
|
148 |
-
name=datasets.Split.TEST,
|
149 |
-
# These kwargs will be passed to _generate_examples
|
150 |
-
gen_kwargs={
|
151 |
-
"filepath": os.path.join(
|
152 |
-
os.path.join(dl_file, "MLQA_V1/test"),
|
153 |
-
f"test-context-{l1}-question-{l2}.json",
|
154 |
-
)
|
155 |
-
},
|
156 |
-
),
|
157 |
-
datasets.SplitGenerator(
|
158 |
-
name=datasets.Split.VALIDATION,
|
159 |
-
# These kwargs will be passed to _generate_examples
|
160 |
-
gen_kwargs={
|
161 |
-
"filepath": os.path.join(
|
162 |
-
os.path.join(dl_file, "MLQA_V1/dev"), f"dev-context-{l1}-question-{l2}.json"
|
163 |
-
)
|
164 |
-
},
|
165 |
-
),
|
166 |
-
]
|
167 |
-
else:
|
168 |
-
if self.config.name.startswith("mlqa-translate-test"):
|
169 |
-
archive = dl_manager.download(self.config.data_url)
|
170 |
-
lang = self.config.name.split(".")[-1]
|
171 |
-
return [
|
172 |
-
datasets.SplitGenerator(
|
173 |
-
name=datasets.Split.TEST,
|
174 |
-
# These kwargs will be passed to _generate_examples
|
175 |
-
gen_kwargs={
|
176 |
-
"filepath": f"mlqa-translate-test/translate-test-context-{lang}-question-{lang}.json",
|
177 |
-
"files": dl_manager.iter_archive(archive),
|
178 |
-
},
|
179 |
-
),
|
180 |
-
]
|
181 |
-
|
182 |
-
def _generate_examples(self, filepath, files=None):
|
183 |
-
"""Yields examples."""
|
184 |
-
if self.config.name.startswith("mlqa-translate"):
|
185 |
-
for path, f in files:
|
186 |
-
if path == filepath:
|
187 |
-
data = json.loads(f.read().decode("utf-8"))
|
188 |
-
break
|
189 |
-
else:
|
190 |
-
with open(filepath, encoding="utf-8") as f:
|
191 |
-
data = json.load(f)
|
192 |
-
for examples in data["data"]:
|
193 |
-
for example in examples["paragraphs"]:
|
194 |
-
context = example["context"]
|
195 |
-
for qa in example["qas"]:
|
196 |
-
question = qa["question"]
|
197 |
-
id_ = qa["id"]
|
198 |
-
answers = qa["answers"]
|
199 |
-
answers_start = [answer["answer_start"] for answer in answers]
|
200 |
-
answers_text = [answer["text"] for answer in answers]
|
201 |
-
yield id_, {
|
202 |
-
"context": context,
|
203 |
-
"question": question,
|
204 |
-
"answers": {"answer_start": answers_start, "text": answers_text},
|
205 |
-
"id": id_,
|
206 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|