ner-wikipedia-dataset.py

from __future__ import annotations

import json
import random
from typing import Generator

from datasets import (
    BuilderConfig,
    DatasetInfo,
    DownloadManager,
    Features,
    GeneratorBasedBuilder,
    Sequence,
    Split,
    SplitGenerator,
    Value,
    Version,
)
from datasets.data_files import DataFilesDict

_CITATION = """
@inproceedings{omi-2021-wikipedia,
    title = "Wikipediaを用いた日本語の固有表現抽出のデータセットの構築",
    author = "近江 崇宏",
    booktitle = "言語処理学会第27回年次大会",
    year = "2021",
    url = "https://anlp.jp/proceedings/annual_meeting/2021/pdf_dir/P2-7.pdf",
}
"""
_DESCRIPTION = "This is a dataset of Wikipedia articles with named entity labels created by Stockmark Inc."
_HOMEPAGE = "https://github.com/stockmarkteam/ner-wikipedia-dataset"
_LICENSE = "CC-BY-SA 3.0"
_URL = "https://raw.githubusercontent.com/stockmarkteam/ner-wikipedia-dataset/main/ner.json"


class NerWikipediaDatasetConfig(BuilderConfig):
    def __init__(
        self,
        name: str = "default",
        version: Version | str | None = Version("0.0.0"),
        data_dir: str | None = None,
        data_files: DataFilesDict | None = None,
        description: str | None = None,
        shuffle: bool = True,
        seed: int = 42,
        train_ratio: float = 0.8,
        validation_ratio: float = 0.1,
    ) -> None:
        super().__init__(
            name=name,
            version=version,
            data_dir=data_dir,
            data_files=data_files,
            description=description,
        )
        self.shuffle = shuffle
        self.seed = seed
        self.train_ratio = train_ratio
        self.validation_ratio = validation_ratio


class NerWikipediaDataset(GeneratorBasedBuilder):
    BUILDER_CONFIG_CLASS = NerWikipediaDatasetConfig

    BUILDER_CONFIGS = [
        NerWikipediaDatasetConfig(
            name="ner-wikipedia-dataset",
            version=Version("2.0.0"),
            description=_DESCRIPTION,
        ),
    ]

    def _info(self) -> DatasetInfo:
        return DatasetInfo(
            description=_DESCRIPTION,
            features=Features(
                {
                    "curid": Value("string"),
                    "text": Value("string"),
                    "entities": [
                        {
                            "name": Value("string"),
                            "span": Sequence(Value("int64"), length=2),
                            "type": Value("string"),
                        }
                    ],
                }
            ),
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(
        self, dl_manager: DownloadManager
    ) -> list[SplitGenerator]:
        dataset_dir = str(dl_manager.download_and_extract(_URL))
        with open(dataset_dir, "r", encoding="utf-8") as f:
            data = json.load(f)

        if self.config.shuffle == True:
            random.seed(self.config.seed)
            random.shuffle(data)

        num_data = len(data)
        num_train_data = int(num_data * self.config.train_ratio)
        num_validation_data = int(num_data * self.config.validation_ratio)
        train_data = data[:num_train_data]
        validation_data = data[
            num_train_data : num_train_data + num_validation_data
        ]
        test_data = data[num_train_data + num_validation_data :]
        return [
            SplitGenerator(
                name=Split.TRAIN,
                gen_kwargs={"data": train_data},
            ),
            SplitGenerator(
                name=Split.VALIDATION,
                gen_kwargs={"data": validation_data},
            ),
            SplitGenerator(
                name=Split.TEST,
                gen_kwargs={"data": test_data},
            ),
        ]

    def _generate_examples(self, data: list[dict[str, str]]) -> Generator:
        for i, d in enumerate(data):
            yield i, {
                "curid": d["curid"],
                "text": d["text"],
                "entities": d["entities"],
            }