ddof parameter on pl.corr is nonsensical

[orlp]

I only realized this after implementing it recently, but the ddof parameter is not used for pl.corr at all. It completely cancels out of the formula. To confirm this in np.corrcoef you'll see:

ddof: _NoValue, optional
    Has no effect, do not use.

We should just remove the parameter in the Rust code and on the Python side deprecate the parameter, removing it in 2.0.

[orlp]

The same holds for pl.rolling_corr.

[zachlefevre]

I am interested in doing this work as my first contribution to Polars.

[zachlefevre]

I'm still working on getting my local dev machine setup -- made more difficult since I use Guix.

[hshreekar]

Hi @orlp, so on the rust side we would need to remove the finalize call or do we need to change the implementation and calculate it using some other formula ?
this is the function pearson_corr in polars/crates/polars-ops/src/chunked_array/cov.rs. I assume this is the implementation and i would need to modify it for every usage of this function and also for Spearman, and the other method in the Enum ?

[orlp]

@hshreekar Don't remove the finalize call, remove the ddof parameter from it. In general from both Spearman and Pearson, not for covariance.

[flowlight0]

Hello! Since other potential contributors looks inactive, let me make my first contributions to Polars :) (#20197)