Integrate Polars library

[albertvillanova]

Check potential integration of the Polars library: /~https://github.com/pola-rs/polars

• Benchmark: https://h2oai.github.io/db-benchmark/

CC: @thomwolf @lewtun

[lewtun]

If possible, a neat API could be something like Dataset.to_polars(), as well as Dataset.set_format("polars")

[albertvillanova]

Note they use a "custom" implementation of Arrow: Arrow2.

[braaannigan]

Polars has grown rapidly in popularity over the last year - could you consider integrating the Polars functionality again?

I don't think the "custom" implementation should be a barrier, it still conforms to the Arrow specification

[amrit110]

Is there some direction regarding this from the HF team @lewtun ? Can conversion from polars to HF dataset be implemented with limited/zero copy? So, something like Dataset.from_polars() and Dataset.to_polars() like you mentioned. Happy to contribute if I can get some pointers on how this may be implemented.

[fzyzcjy]

Hi, is there any updates? Thanks!

[baggiponte]

Hi, is there any updates? Thanks!

The feature has been there for a bit 😊 You can call dataset.to_polars() (on a Dataset, not a DatasetDict). The issue can be closed, I guess! @lhoestq

[fzyzcjy]

Looks great and thanks!

[albertvillanova]

Thank you.