import json
import datasets

class SanskritWordSegmentation(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")

    def _info(self):
        return datasets.DatasetInfo(
            description="Gold-standard Sanskrit word segmentation and morphological candidate graphs.",
            features=datasets.Features({
                "id": datasets.Value("string"),
                "sentence": datasets.Value("string"),
                "gold_segments": datasets.Sequence(datasets.Value("string")),
                "lemmas": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
                "morph_tags": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
                "candidates": datasets.Sequence({
                    "id": datasets.Value("string"),
                    "word": datasets.Value("string"),
                    "lemma": datasets.Value("string"),
                    "morph": datasets.Value("string"),
                    "cng": datasets.Value("string"),
                    "chunk_no": datasets.Value("string"),
                    "position": datasets.Value("int32"),
                    "length": datasets.Value("int32"),
                    "pre_verb": datasets.Value("string"),
                }),
            }),
            supervised_keys=None,
            homepage="https://zenodo.org/records/803508",
            citation="Krishna et al. (2017), A Dataset for Sanskrit Word Segmentation",
        )

    def _split_generators(self, dl_manager):
        data_path = dl_manager.download_and_extract("data.jsonl")
        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_path}),
        ]

    def _generate_examples(self, filepath):
        with open(filepath, encoding="utf-8") as f:
            for idx, line in enumerate(f):
                yield idx, json.loads(line)