WRENCH.py

"""WRENCH classification dataset."""

import json

import datasets


class WrenchClassifConfig(datasets.BuilderConfig):
    """BuilderConfig for WRENCH."""

    def __init__(self, dataset_path, **kwargs):
        super(WrenchClassifConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)
        self.dataset_path = dataset_path


class WrenchSequenceTaggingConfig(datasets.BuilderConfig):
    """BuilderConfig for WRENCH."""

    def __init__(self, dataset_path, **kwargs):
        super(WrenchClassifConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)
        self.dataset_path = dataset_path


class Wrench(datasets.GeneratorBasedBuilder):
    """WRENCH classification dataset."""

    ENTITY_RELATED_DATASETS = ["spouse", "cdr", "semeval", "chemprot"]

    SEQUENCETAGGING_DATASET = [
        {"name": "conll", "dataset_path": "./conll"},
        {"name": "laptopreview", "dataset_path": "./laptopreview"},
        {"name": "bc5cdr", "dataset_path": "./bc5cdr"},
        {"name": "mit-movies", "dataset_path": "./mit-movies"},
        {"name": "mit-restaurants", "dataset_path": "./mit-restaurants"},
        {"name": "ncbi-disease", "dataset_path": "./ncbi-disease"},
        {"name": "ontonotes", "dataset_path": "./ontonotes"},
        {"name": "wikigold", "dataset_path": "./wikigold"},
    ]

    CLASSIF_DATASETS = [
        {"name": "imdb", "dataset_path": "./imdb"},
        {"name": "agnews", "dataset_path": "./agnews"},
        {"name": "cdr", "dataset_path": "./cdr"},
        {"name": "chemprot", "dataset_path": "./chemprot"},
        {"name": "semeval", "dataset_path": "./semeval"},
        {"name": "sms", "dataset_path": "./sms"},
        {"name": "spouse", "dataset_path": "./spouse"},
        {"name": "trec", "dataset_path": "./trec"},
        {"name": "yelp", "dataset_path": "./yelp"},
        {"name": "youtube", "dataset_path": "./youtube"},
    ]

    configs_clf = [
        WrenchClassifConfig(name=i["name"], dataset_path=i["dataset_path"]) for i in CLASSIF_DATASETS
    ]
    BUILDER_CONFIGS = configs_clf

    def _info(self):
        name = self.config.name
        if name in self.ENTITY_RELATED_DATASETS:
            return datasets.DatasetInfo(
                features=datasets.Features(
                    {
                        "text": datasets.Value("string"),
                        "label": datasets.Value("int8"),
                        "entity1": datasets.Value("string"),
                        "entity2": datasets.Value("string"),
                        "weak_labels": datasets.Sequence(datasets.Value("int8")),
                    }
                )
            )
        else:
            return datasets.DatasetInfo(
                features=datasets.Features(
                    {
                        "text": datasets.Value("string"),
                        "label": datasets.Value("int8"),
                        "weak_labels": datasets.Sequence(datasets.Value("int8")),
                    }
                )
            )

    def _split_generators(self, dl_manager):
        dataset_path = self.config.dataset_path
        name = self.config.name
        train_path = dl_manager.download_and_extract(f"{dataset_path}/train.json")
        valid_path = dl_manager.download_and_extract(f"{dataset_path}/valid.json")
        test_path = dl_manager.download_and_extract(f"{dataset_path}/test.json")
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, gen_kwargs={"filepath": train_path, "name": name}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION, gen_kwargs={"filepath": valid_path, "name": name}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST, gen_kwargs={"filepath": test_path, "name": name}
            ),
        ]

    def _generate_examples(self, filepath, name):
        """Generate Custom examples."""

        with open(filepath, encoding="utf-8") as f:
            json_data = json.load(f)

        if name in self.ENTITY_RELATED_DATASETS:
            for idx in json_data:
                data = json_data[idx]

                text = data["data"]["text"]
                weak_labels = data["weak_labels"]
                label = data["label"]
                entity1 = data["data"]["entity1"]
                entity2 = data["data"]["entity2"]

                yield int(idx), {
                    "text": text,
                    "label": label,
                    "entity1": entity1,
                    "entity2": entity2,
                    "weak_labels": weak_labels,
                }
        else:
            if name == "trec":
                for idx in json_data:
                    data = json_data[idx]
                    text = " ".join(data["data"]["text"].split(" ")[1:])
                    weak_labels = data["weak_labels"]
                    label = data["label"]

                    yield int(idx), {
                        "text": text,
                        "label": label,
                        "weak_labels": weak_labels,
                    }
            else:
                for idx in json_data:
                    # print(json_data[idx])
                    data = json_data[idx]

                    text = data["data"]["text"]
                    weak_labels = data["weak_labels"]
                    label = data["label"]

                    yield int(idx), {
                        "text": text,
                        "label": label,
                        "weak_labels": weak_labels,
                    }