tfds.load() does not load datasets with a capital letter

[BogueUser]

Short description

Running tfds build Mk0_datasets_builder.py will save to ~/tensorflow_datasets/Mk0
When running tfds.load('Mk0', split='train', shuffle_files=True) to import it, the following error is given.

No registered data_dirs were found in:
        - /home/user/tensorflow_datasets

Renaming the file to mk0 from Mk0 will allow it to load however.

Environment information

• Operating System: Arch Linux

• Python version: 3.11.5

• `tensorflow-datasets version: 4.9.4

• tensorflow version: 2.14.0

• Does the issue still exists with the last tfds-nightly package (pip install --upgrade tfds-nightly) ?

• Yes

Reproduction instructions
Build a dataset with a capital letter in the name then attempt to load with tfds

tfds.load('Mk0', split='train', shuffle_files=True)

Expected behavior
Either tfds build should automatically make the name lowercase or tfds.load() should be able to deal with uppercase letters

[fineguy]

Thanks for reporting this issue!

This is indeed a real problem, we'll need to think if supporting uppercase in tfds.load is possible. In the meanwhile you should stick with lowercase for your dataset names. Sorry for the inconvenience.

[BogueUser]

Awesome. Thank you for looking into this.