germandpr-beir.py

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 and the other dataset is preprocessed. See official dataset card for " \
               "usage of dataset with BEIR."

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


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

    DEFAULT_CONFIG_NAME = "qrels"

    def _info(self):
        name = self.config.name
        _SPLITS = ["queries-original", "corpus-original", "queries-processed", "corpus-processed", "qrels"]

        if name.startswith("queries"):
            features = {
                "_id": datasets.Value("string"),
                "text": datasets.Value("string")
            }
        elif name.startswith("corpus"):
            features = {
                "_id": datasets.Value("string"),
                "title": datasets.Value("string"),
                "text": datasets.Value("string"),
            }
        else:
            # name == qrels
            features = {
                "query-id": datasets.Value("string"),
                "corpus-id": datasets.Value("string"),
                "score": datasets.Value("int32")
            }

        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."""
        _SPLITS = ["queries-original", "corpus-original", "queries-processed", "corpus-processed", "qrels"]

        name = self.config.name
        if name == "qrels":
            dl_path = dl_manager.download([
                "https://huggingface.co/datasets/PM-AI/germandpr-beir/resolve/main/data/qrels/test.tsv",
                "https://huggingface.co/datasets/PM-AI/germandpr-beir/resolve/main/data/qrels/train.tsv"
            ])
        else:
            dl_path = dl_manager.download(f"https://huggingface.co/datasets/PM-AI/germandpr-beir/resolve/main/data/{name}.jsonl")

        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": dl_path, "split": "train"}),
            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": dl_path, "split": "test"})
        ]

    def _generate_examples(self, filepath, split):
        """Yields examples."""
        name = self.config.name
        if name.startswith("queries"):
            yield 0, {"_id": "1", "text": "text"}
        elif name.startswith("corpus"):
            yield 0, {"_id": "1", "title": "title", "text": "text"}
        else:
            # name == qrels
            filepath = [x for x in filepath if x.endswith(f"{split}.tsv")]
            yield 0, {"query-id": "", "corpus-id": "", "score": 1}