commavq.py

import datasets
import glob
import os
import numpy as np

SHARD_SIZE = 2500
NUM_SHARDS = 40
_URLS = [
  f'https://huggingface.co/datasets/commaai/commavq/resolve/main/data_{i*SHARD_SIZE}_to_{(i+1)*SHARD_SIZE}.zip' for i in range(NUM_SHARDS)
  ]

_DESCRIPTION = """\
TODO
"""

class CommaVQ(datasets.GeneratorBasedBuilder):

  def _info(self):
    return datasets.DatasetInfo(
      description=_DESCRIPTION,
      features=datasets.Features(
        {"path": datasets.Value("string")}
      )
    )

  def _split_generators(self, dl_manager):
    """Returns SplitGenerators."""
    downloaded_files = dl_manager.download_and_extract(_URLS)
    return [datasets.SplitGenerator(name=str(i), gen_kwargs={"filepaths": x}) for i, x in enumerate(downloaded_files)]

  def _generate_examples(self, filepaths):
    for path in glob.glob(f'{filepaths}/*.npy'):
      name = os.path.basename(path)
      yield name, {'path': path}