scikit-learn-contrib
Native support for `polars` dataframes in imbalanced-learn
Polars is a high-performance DataFrame library for Python, celebrated for its fast data processing capabilities and efficient, concise syntax. Its multi-threaded query engine and strong integration with the Python ecosystem make it an outstanding choice for managing large datasets. Polars has been gaining popularity as a fast and memory-efficient alternative to pandas, especially for big data applications.
While several libraries like scikit-learn and seaborn have added support for Polars DataFrames, I am not sure if imbalanced-learn currently requires users to directly use polars dataFrames or they need to convert Polars DataFrames to pandas (e.g., polars_df.to_pandas()) before applying the sampling methods.
I do see that imblearn depends upon polars and some APIs like set_output accept polars as parameter, it's unclear if imblearn APIs can directly work with polars dataframes
Maybe someone could point out the necessary steps to get at least rudimentary functionality with polars dataframes. I would be willing to investigate implementing those. Thanks!
In case someone else is looking to use a custom sampler, FunctionSampler(..., validate=False) allows any data type as described in its docs. This allows polars inputs, but obviously won't validate them as it does polars/numpy
