Does a keep_columns method on Dataset, IterableDataset, etc. make sense?

[alvarobartt]

I was dealing with a Dataset with way too many columns and I feel a bit tedious to do a set(columns) - columns_to_keep to then provide that as a list to .remove_columns().

AFAIK this has been mentioned by other users in the past e.g. https://discuss.huggingface.co/t/how-to-create-a-new-dataset-from-another-dataset-and-select-specific-columns-and-the-data-along-with-the-column/15120, so does it make sense to add a method named .keep_columns() or something similar to do that, so that the user doesn't need to calculate the column diff on top of .remove_columns().

I also understand that this may be a little over-engineering for such a small feature that maybe is not that common and in the end, it's just helping the developer with removing a single line of code, but I think it can help somehow. Anyway, I also get the counterpart engineering-wise.

So I'd appreciate your input here!

[mariosasko]

Hi! We are in the process of adding this method to the API (see #5468).

[alvarobartt]

Hi, @mariosasko thanks for the follow-up, I'll close this Discussion when merged into main if you agree 🤗

[alvarobartt]

Thanks for that @mariosasko! To anyone else reading, PR at #5480 was merged yesterday and is already available if you install datasets from source as pip install git+https://github.com/huggingface/datasets@69b19755e9e37b746ef56780a62d21ef20c574d5