Apache arrow support for I/O

[Trollgeir]

catboost version: 1.03 Operating System: ubuntu linux

I'm using the new and promising library Polars for transforming my data, and my goal is to have as little memory impact as possible because of constraints. Would it be possible to support the apache arrow data format as an input for catboost? The underlying dictionary types fit perfect for categorical features.

[braaannigan]

+1 for this

XGBoost added support for Arrow ingest a few months ago here: https://github.com/dmlc/xgboost/pull/7512

[jcierocki]

This would be fantastic! LightGBM has this functionality available for pretty long time now.

[arrowcircle]

+1

[anthonygiorgio97]

+1

It would be great for polars support

[aryehklein-rise]

+1