my_npz_dataset.py

import os
import numpy as np
from datasets import Dataset, DatasetDict, Features, Array3D, Value, GeneratorBasedBuilder

class MyNpzDatasetBuilder(GeneratorBasedBuilder):
    def _info(self):
        features = Features({
            'sdf': Array3D(dtype='float64', shape=(512, 512)),  # Adjust dtype and shape according to your data
            'mask': Array3D(dtype='int64', shape=(512, 512)),
            're': Array3D(dtype='float64', shape=(512, 512)),
            'u': Array3D(dtype='float64', shape=(512, 512)),
            'v': Array3D(dtype='float64', shape=(512, 512)),
            'p': Array3D(dtype='float64', shape=(512, 512)),
        })
        return datasets.DatasetInfo(
            description="Flow Bench Dataset for SciML",
            features=features,
            supervised_keys=None,
        )

    def _split_generators(self, dl_manager):
        data_dir = dl_manager.download_and_extract(self.config.data_files)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"file_path": os.path.join(data_dir, "lid_driven_cavity_NS_512x512_X.npz")}
            )
        ]

    def _generate_examples(self, file_path):
        data = np.load(file_path)
        
        for idx in range(len(data['sdf'])):
            yield idx, {
                'sdf': data['sdf'][idx],
                'mask': data['mask'][idx],
                're': data['re'][idx],
                'u': data['u'][idx],
                'v': data['v'][idx],
                'p': data['p'][idx],
            }

# Example usage
if __name__ == "__main__":
    from datasets import load_dataset

    dataset = load_dataset("./my_npz_dataset.py", data_files={"train": "./fake_lid_driven_cavity_NS_512x512_X.npz"})
    print(dataset)