fever-span-annotated.py

import gzip
import pandas as pd
import datasets
from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split
import csv
from itertools import islice

class FeverDataset(GeneratorBasedBuilder):
    VERSION = "1.0.0"

    def _info(self):
        return DatasetInfo(
            description="FEVER Dataset with span annotations.",
            features=datasets.Features({
                "query": datasets.Value("string"),
                "document": datasets.Value("string"),
                "spans": datasets.Value("string"),
            }),
            supervised_keys=None,
            homepage="https://huggingface.co/datasets/jinaai/fever-span-annotated",
            license="CC BY 4.0",
            splits={
                "test": {"num_bytes": 0},   # Placeholder, will be updated later
            },
            dataset_size=0,
            download_size=0
        )

    def _download_and_prepare(self, dl_manager, **kwargs):
        # Define the URLs for the dataset files
        urls = [
            "https://huggingface.co/datasets/jinaai/fever-span-annotated/resolve/main/data/fever-pairs_shard_000000.tsv.gz",
            "https://huggingface.co/datasets/jinaai/fever-span-annotated/resolve/main/data/fever-pairs_shard_000001.tsv.gz",
            "https://huggingface.co/datasets/jinaai/fever-span-annotated/resolve/main/data/fever-pairs_shard_000002.tsv.gz",
            "https://huggingface.co/datasets/jinaai/fever-span-annotated/resolve/main/data/fever-pairs_shard_000003.tsv.gz",
            "https://huggingface.co/datasets/jinaai/fever-span-annotated/resolve/main/data/fever-pairs_shard_000004.tsv.gz",
        ]

        # Download the files using the dl_manager
        downloaded_files = dl_manager.download(urls)

        # Set the file paths for the split generators
        self.file_paths = downloaded_files

    def _split_generators(self, dl_manager):
        return [
            SplitGenerator(
                name=Split.TEST,
                gen_kwargs={"file_paths": self.file_paths},
            ),
        ]

    def _generate_examples(self, file_paths):
        for file_path in file_paths:
            with gzip.open(file_path, 'rt') as file:
                reader = csv.reader(
                    file,
                    dialect='excel-tab' if self._dialect == 'tsv' else 'excel',
                )
                for row in islice(reader, self._current_index, None, self._stride):
                    if len(row) >= 3:
                        yield row[0], {
                            "query": row[0],
                            "document": row[1],
                            "spans": row[2],
                        }