`from_generator` does not allow to specify the split name

[pminervini]

Describe the bug

I'm building train, dev, and test using from_generator; however, in all three cases, the logger prints Generating train split: It's not possible to change the split name since it seems to be hardcoded: https://github.com/huggingface/datasets/blob/main/src/datasets/packaged_modules/generator/generator.py

Steps to reproduce the bug

In [1]: from datasets import Dataset

In [2]: def gen():
   ...:     yield {"pokemon": "bulbasaur", "type": "grass"}
   ...: 

In [3]: ds = Dataset.from_generator(gen)
Generating train split: 1 examples [00:00, 133.89 examples/s]

Expected behavior

It should be possible to specify any split name

Environment info

• datasets version: 2.19.2
• Platform: macOS-10.16-x86_64-i386-64bit
• Python version: 3.8.5
• huggingface_hub version: 0.23.3
• PyArrow version: 15.0.0
• Pandas version: 2.0.3
• fsspec version: 2023.10.0

[albertvillanova]

Thanks for reporting, @pminervini.

I agree we should give the option to define the split name.

Indeed, there is a PR that addresses precisely this issue:

• 7015

I am reviewing it.

[pminervini]

Booom! thank you guys :)