parashoot.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import io
import json
import os

import datasets
from datasets.tasks import QuestionAnsweringExtractive


logger = datasets.logging.get_logger(__name__)


_CITATION = """\
@inproceedings{keren2021parashoot,
  title={ParaShoot: A Hebrew Question Answering Dataset},
  author={Keren, Omri and Levy, Omer},
  booktitle={Proceedings of the 3rd Workshop on Machine Reading for Question Answering},
  pages={106--112},
  year={2021}
}
"""

_DESCRIPTION = """
A Hebrew question and answering dataset in the style of SQuAD, based on articles scraped from Wikipedia. The dataset contains a few thousand crowdsource-annotated pairs of questions and answers, in a setting suitable for few-shot learning.
"""

_URLS = {
    "train": "data/train.tar.gz",
    "validation": "data/dev.tar.gz",
    "test": "data/test.tar.gz",
}


class ParashootConfig(datasets.BuilderConfig):
    """BuilderConfig for Parashoot."""

    def __init__(self, **kwargs):
        """BuilderConfig for Parashoot.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(ParashootConfig, self).__init__(**kwargs)


class Parashoot(datasets.GeneratorBasedBuilder):
    """Parashoot: The Hebrew Question Answering Dataset. Version 1.1."""

    BUILDER_CONFIGS = [
        ParashootConfig(
            version=datasets.Version("1.1.0", ""),
            description=_DESCRIPTION,
        ),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "id": datasets.Value("string"),
                    "title": datasets.Value("string"),
                    "context": datasets.Value("string"),
                    "question": datasets.Value("string"),
                    "answers": datasets.features.Sequence(
                        {
                            "text": datasets.Value("string"),
                            "answer_start": datasets.Value("int32"),
                        }
                    ),
                }
            ),
            # No default supervised_keys (as we have to pass both question
            # and context as input).
            supervised_keys=None,
            homepage="https://github.com/omrikeren/ParaShoot",
            citation=_CITATION,
            task_templates=[
                QuestionAnsweringExtractive(
                    question_column="question",
                    context_column="context",
                    answers_column="answers",
                )
            ],
        )

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

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": dl_manager.iter_archive(downloaded_files["train"]),
                    "basename": "train.jsonl",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "filepath": dl_manager.iter_archive(downloaded_files["validation"]),
                    "basename": "dev.jsonl",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "filepath": dl_manager.iter_archive(downloaded_files["test"]),
                    "basename": "test.jsonl",
                },
            ),
        ]

    def _generate_examples(self, filepath, basename):
        """This function returns the examples in the raw (text) form."""
        logger.info("generating examples from = %s", filepath)
        key = 0
        for file_path, file_obj in filepath:
            with io.BytesIO(file_obj.read()) as f:
                for line in f:
                    article = json.loads(line)
                    title = article.get("title", "")
                    context = article["context"]
                    answer_starts = article["answers"]["answer_start"]
                    answers = article["answers"]["text"]
                    yield key, {
                        "title": title,
                        "context": context,
                        "question": article["question"],
                        "id": article["id"],
                        "answers": {
                            "answer_start": answer_starts,
                            "text": answers,
                        },
                    }
                    key += 1