import json
import datasets

class ArxivTables(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")

    def _info(self):
        return datasets.DatasetInfo(
            description="Arxiv-tables Dataset",
            features=datasets.Features({
                "id": datasets.Value("int32"),
                "page": datasets.Value("int32"),
                "bounding_box": datasets.Sequence(datasets.Value("float32"), length=4),
                "latex_content": datasets.Value("string"),
                "extracted_content": datasets.Value("string"),
                "similarity_score": datasets.Value("float32"),
                "table_image": datasets.Image(),
                "page_image": datasets.Image(),
            }),
        )

    def _split_generators(self, dl_manager):
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"path": "arxiv_tables"},
            ),
        ]

    def _generate_examples(self, path):
        for arxiv_id in os.listdir(path):
            arxiv_path = os.path.join(path, arxiv_id)
            for table_dir in os.listdir(arxiv_path):
                table_path = os.path.join(arxiv_path, table_dir)
                with open(os.path.join(table_path, "metadata.json"), "r") as f:
                    metadata = json.load(f)
                
                yield metadata["id"], {
                    "id": metadata["id"],
                    "page": metadata["page"],
                    "bounding_box": metadata["bounding_box"],
                    "latex_content": metadata["latex_content"],
                    "extracted_content": metadata["extracted_content"],
                    "similarity_score": metadata["similarity_score"],
                    "table_image": os.path.join(table_path, "table_image.png"),
                    "page_image": os.path.join(table_path, "page_image.png"),
                }