Error "Nested data conversions not implemented for chunked array outputs"

#2
by ShuokaiPan - opened

HI there,

Thanks a lot for releasing this dataset.
I was trying to load this dataset like the following, but was seeing this error
ArrowNotImplementedError: Nested data conversions not implemented for chunked array outputs

from datasets import load_dataset
dataset = load_dataset('guangyil/laion-coco-aesthetic', split='train')

Can you let me know where I did wrong here? Thanks a lot for your help.
Cheers

Is it because there is only one shard for this dataset and this one shard is too large?

I didn't split it into train/val/test. I think you can try the command directly. Or, you may git clone the repo and use pandas to load the parquet file.
dataset = load_dataset("guangyil/laion-coco-aesthetic")

Hi,

Thanks for the reply.
I tried to load it like this dataset = load_dataset("guangyil/laion-coco-aesthetic"), still have the same error.
When I tried to load using pandas like this pd.read_parquet('./laion-coco_v3_filter.parquet'), still the same error.
But this works load_dataset('guangyil/laion-coco-aesthetic', revision="refs/convert/parquet"). In that branch, there are multiple smaller parquet files.

So maybe this is indeed to do with one large shard file?
Cheers

Sign up or log in to comment