Error in the loading script
Continuation of the Issue discussion in GitHub: https://github.com/huggingface/datasets/issues/4443
We can use this discussion page to raise errors we find in the script and ask/answer/propose solutions.
After fixing the previous KeyError, now there appears a NotFoundError error:
Exception: NotFoundError
Message: 20200718/train/random_crops_256/seq-24-00000-of-02100.tfrecord.gz; No such file or directory [Op:IteratorGetNext]
The script cannot find the file "20200718/train/random_crops_256/seq-24-00000-of-02100.tfrecord.gz", specified as a relative path in the current repository.
Yeah, I believe that is from the download(_URL) downloading the json file of all the paths, but not the actual data files
Note that the dataset viewer is currently failing to extract the first 100 rows of every split because it runs out of memory (it can get up to ~32GB of RAM).
Ah okay, I'm not sure if I can get it that much smaller, it is just fairly large examples, I think.
Does Datasets support reading from hosted TFRecord files? For the non-sample files (so the files hosted in the repo) the Dataset viewer is now giving the error
File system scheme 'https' not implemented (file: 'https://huggingface.co/datasets/openclimatefix/nimrod-uk-1km/resolve/main/20200718/train/random_crops_256/seq-24-00000-of-02100.tfrecord.gz') [Op:IteratorGetNext
I'm assuming the dataset viewer uses the dataset in streaming mode? Is there a way to support TFRecord files? Would un-gzipping the files and reuploading them work?
Maybe @lhoestq @mariosasko @albertvillanova could help
I'm currently converting it to Parquet slowly and uploading those which seems to work well and is easy enough, just takes awhile
HI! I also encountered the same mistake. I don't know how to solve it ---> 20200718/train/random_crops_256/seq-24-00000-of-02100.tfrecord.gz