triviaqa-squad-web-br.py

"""TODO: DATASET SHORT DESCRIPTION"""

import json
import datasets
from datasets.tasks import TextClassification

_DESCRIPTION = """TODO: DATASET DESCRIPTION"""

_CITATION = """TODO: CITATIONS"""

#TODO: official mirror url
_URL = "https://huggingface.co/datasets/rsgrava/triviaqa-squad-web-br/resolve/main/"
_URLS = {
    "web-train": _URL + "web-train-squad-br.json",
    "web-dev": _URL + "web-dev-squad-br.json",
}

class SquadV1PtBr(datasets.GeneratorBasedBuilder):
    """TODO: SHORT DATASET DESCRIPTION"""

    VERSION = datasets.Version("1.1.0")

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "id": datasets.Value("string"),
                    "context": datasets.Value("string"),
                    "qas": datasets.features.Sequence(
                        {
                            "question": datasets.Value("string"),
                            "answers": datasets.features.Sequence(
                                {
                                    "text": datasets.Value("string"),
                                }
                            )
                        }
                    ),
                }
            ),
            # No default supervised_keys (as we have to pass both question and context as input).
            supervised_keys=None,
            homepage="TODO: HOMEPAGE",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        urls_to_download = _URLS
        downloaded_files = dl_manager.download_and_extract(urls_to_download)

        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["web-train"]}),
            datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["web-dev"]}),
        ]

    def _generate_examples(self, filepath):
        with open(filepath, 'r') as f:
            key = 0
            ids = []
            squad = [json.loads(line) for line in f.read().splitlines()]
            for example in squad:
                context = example["context"].strip()
                id_ = example["id"]
                if id_ in ids:
                    #skip duplicate entries
                    continue
                else:
                    ids.append(id_)
                qas = []
                qa = example["qas"]
                answers = []
                for answer in qa["answers"]:
                    answers.append({
                        "text": answer.strip(),
                    })
                qas.append({
                    "question": qa["question"].strip(),
                    "answers":  answers
                })
                yield key, {
                    "id": key,
                    "context": context,
                    "qas": qas,
                }
                key += 1