germandpr-beir.py

import json
import os.path

import datasets


_VERSION = "1.0.0"

_DESCRIPTION = "Deepset's germanDPR dataset made compatible with BEIR benchmark framework. One version contains " \
               "the original dataset 1:1 (but deduplicated) and the other dataset is furhter preprocessed. " \
               "See official dataset card for dataset usage with BEIR."

_SUBSETS = ["original-queries", "original-corpus", "original-qrels",
            "processed-queries", "processed-corpus", "original-qrels"]


class GermanDPRBeir(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = (
        [
            datasets.BuilderConfig(
                name="original-queries",
                description=f"BEIR queries created 1:1 but deduplicated from deepset/germanDPR.",
                version=_VERSION,
            ),
            datasets.BuilderConfig(
                name="original-corpus",
                description=f"BEIR corpus created 1:1 but deduplicated from deepset/germanDPR.",
                version=_VERSION,
            ),
            datasets.BuilderConfig(
                name="original-qrels",
                description=f"BEIR qrels for original version of deepset/germanDPR.",
                version=_VERSION,
            ),
            datasets.BuilderConfig(
                name="processed-queries",
                description=f"BEIR queries created, deduplicated and further text-processed from deepset/germanDPR.",
                version=_VERSION,
            ),
            datasets.BuilderConfig(
                name="processed-corpus",
                description=f"BEIR corpus created, deduplicated and further text-processed from deepset/germanDPR.",
                version=_VERSION,
            ),
            datasets.BuilderConfig(
                name="processed-qrels",
                description=f"BEIR qrels for processed version of deepset/germanDPR.",
                version=_VERSION,
            )
        ]
    )

    DEFAULT_CONFIG_NAME = _SUBSETS[0]

    def _info(self):
        name = self.config.name
        if name.endswith("queries"):
            features = {
                "_id": datasets.Value("string"),
                "text": datasets.Value("string")
            }
        elif name.endswith("corpus"):
            features = {
                "_id": datasets.Value("string"),
                "title": datasets.Value("string"),
                "text": datasets.Value("string"),
            }
        elif name.endswith("qrels"):
            # name == qrels
            features = {
                "query-id": datasets.Value("string"),
                "corpus-id": datasets.Value("string"),
                "score": datasets.Value("int32")
            }
        else:
            raise ValueError(f'Unknown subset, choose from: {", ".join(_SUBSETS)}')

        return datasets.DatasetInfo(
            description=f"{_DESCRIPTION}\n{self.config.description}",
            features=datasets.Features(features),
            supervised_keys=None,
            homepage="https://huggingface.co/datasets/PM-AI/germandpr-beir",
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        name = self.config.name
        if name.startswith("original"):
            dl_path = dl_manager.download_and_extract("https://huggingface.co/datasets/PM-AI/germandpr-beir/resolve/main/data/original.tar.gz")
        elif name.startswith("processed"):
            dl_path = dl_manager.download_and_extract("https://huggingface.co/datasets/PM-AI/germandpr-beir/resolve/main/data/processed.tar.gz")
        else:
            raise ValueError(f'Unknown subset, choose from: {", ".join(_SUBSETS)}')

        type1, type2 = name.split("-")
        if type2 in ["corpus", "queries"]:
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN,
                    gen_kwargs={"filepath": os.path.join(dl_path, f'{type1}/train/{type2}.jsonl')}),
                datasets.SplitGenerator(
                    name=datasets.Split.TEST,
                    gen_kwargs={"filepath": os.path.join(dl_path, f'{type1}/test/{type2}.jsonl')})
            ]
        elif type2 == "qrels":
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN,
                    gen_kwargs={"filepath": os.path.join(dl_path, f'{type1}/train/qrels/train.tsv')}),
                datasets.SplitGenerator(
                    name=datasets.Split.TEST,
                    gen_kwargs={"filepath": os.path.join(dl_path, f'{type1}/test/qrels/test.tsv')})
            ]
        else:
            raise ValueError(f'Unknown subset, choose from: {", ".join(_SUBSETS)}')

    def _generate_queries_data(self, filepath):
        print("filepath: ", filepath)
        with open(filepath, "r", encoding="utf-8") as in_file:
            for idx, line in enumerate(in_file):
                data = json.loads(line)
                yield idx, data

    def _generate_corpus_data(self, filepath):
        with open(filepath, "r", encoding="utf-8") as in_file:
            for idx, line in enumerate(in_file):
                data = json.loads(line)
                if "metadata" in data:
                    del data["metadata"]
                yield idx, data

    def _generate_qrel_data(self, filepath):
        with open(filepath, "r", encoding="utf-8") as in_file:
            in_file.readline() # first line is header
            for idx, line in enumerate(in_file):
                qid, cid, score = line.rstrip().split("\t")
                yield idx, {"query-id": qid, "corpus-id": cid, "score": score}

    def _generate_examples(self, filepath):
        """Yields examples."""
        name = self.config.name
        if name.endswith("queries"):
            return self._generate_queries_data(filepath)
        elif name.endswith("corpus"):
            return self._generate_corpus_data(filepath)
        elif name.endswith("qrels"):
            return self._generate_qrel_data(filepath)
        else:
            raise ValueError(f'Unknown subset, choose from: {", ".join(_SUBSETS)}')