semi-heter.py

import os
import datasets
import pandas as pd
import json


class semiHeterConfig(datasets.BuilderConfig):
    def __init__(self, features, data_url, **kwargs):
        super(semiHeterConfig, self).__init__(**kwargs)
        self.features = features
        self.data_url = data_url


class semiHeter(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        semiHeterConfig(
            name="pairs",
            features={
                "ltable_id": datasets.Value("string"),
                "rtable_id": datasets.Value("string"),
                "label": datasets.Value("string"),
            },
            data_url="https://huggingface.co/datasets/matchbench/semi-heter/resolve/main/",
        ),
        semiHeterConfig(
            name="source",
            features={
                "content": datasets.Value("string"),
            },
            data_url="https://huggingface.co/datasets/matchbench/semi-heter/resolve/main/left.json",
        ),

        semiHeterConfig(
            name="target",
            features={
                "content": datasets.Value("string"),
            },
            data_url="https://huggingface.co/datasets/matchbench/semi-heter/resolve/main/right.json",
        ),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            features=datasets.Features(self.config.features)
        )

    def _split_generators(self, dl_manager):
        if self.config.name == "pairs":
            return [
                datasets.SplitGenerator(
                    name=split,
                    gen_kwargs={
                        "path_file": dl_manager.download_and_extract(
                            os.path.join(self.config.data_url, f"{split}.csv")),
                        "split": split,
                    }
                )
                for split in ["train", "valid", "test"]
            ]

        if self.config.name == "source":
            return [datasets.SplitGenerator(name="source", gen_kwargs={
                "path_file": dl_manager.download_and_extract(self.config.data_url), "split": "source", })]

        if self.config.name == "target":
            return [datasets.SplitGenerator(name="target", gen_kwargs={
                "path_file": dl_manager.download_and_extract(self.config.data_url), "split": "target", })]

    def _generate_examples(self, path_file, split):
        if split in ['source', 'target']:
            with open(path_file, "r") as f:
                file = json.load(f)
                for i in range(len(file)):
                    yield i, {
                        "content": json.dumps(file[i])
                    }
        else:
            file = pd.read_csv(path_file)
            for i, row in file.iterrows():
                    yield i, {
                        "ltable_id": row["ltable_id"],
                        "rtable_id": row["rtable_id"],
                        "label": row["label"],
                    }