DropSimilar requires pyarrow to be installed when the input is a polars dataframe, because compute_associations uses pandas to compute associations. However, DropSimilar does not use the Pearson correlations (which is what actually requires pyarrow). compute_associations should have a parameter to compute either Cramer V, Pearson or both (both by default), and DropSimilar should skip Pearson, since it's not used.
We may also want to use that option in the tablereport and compute only cramerv with a warning when we have a polars dataframe and no pyarrow.
DropSimilar requires pyarrow to be installed when the input is a polars dataframe, because compute_associations uses pandas to compute associations. However, DropSimilar does not use the Pearson correlations (which is what actually requires pyarrow).
compute_associationsshould have a parameter to compute either Cramer V, Pearson or both (both by default), and DropSimilar should skip Pearson, since it's not used.We may also want to use that option in the tablereport and compute only cramerv with a warning when we have a polars dataframe and no pyarrow.