Datasets:

commaai

/

commavq

like 11

Follow

comma.ai 17

Hi ! videos are not supported yet by the dataset viewer, sorry for the inconvenience

This dataset is only strings (paths to numpy arrays)

https://huggingface.co/datasets/commaai/commavq/blob/main/commavq.py#L22

It seems that you override_get_examples_iterable_for_split while you should only implement _split_generators and _generate_examples. Doing so can lead to unexpected behaviors. Please refer to the documentation about how to create a dataset script.

Your code is not supported by the dataset viewer.

Let me open a PR to make your code not rely on overriding _get_examples_iterable_for_split if you don't mind

I just opened https://huggingface.co/datasets/commaai/commavq/discussions/2, let me know what you think.

Also may I ask why there is one split per shard ? Usually splits are for train/validation/tests so I'm curious what's your use case.

Thanks for the fix!
I did 1 split per shard in order to have the possibility to download only 1 or a few shards.
AFAIK this can't be done without using splits

dataset = load_dataset("commaai/commavq", num_proc=num_proc_load_dataset, split=['0', '1'])

I am closing this viewer issue as already fixed.

In relation with the question about the usage, @Yassine please note that for a specific configuration, ALL splits are downloaded and cached, but only those passed in the split parameter are loaded (i.e. memory-mapped).

FYI usually we define splits as train/validation/tests and we use another argument to specify which shards to load, e.g. data_files (that you can get in your dataset script as self.config.data_files)

Thanks for letting me know! @lhoestq I thought only the specified splits were downloaded.
Will look into the data_files!