Add rsquared, pvalue, etc to results

[jackaixin]

Hi @azmyrajab, I’m really a fan of this extension, which speeds up my cross-sectional regression task by more than 10x (i was using statsmodels before). However, I’m wondering if it’s possible to add other metrics that are available in statsmodels’ results. Common metrics include rsquared, rsquared_adj, pvalue, tvalue, fvalue, nobs, etc. Maybe there can be an argument specifying which metrics are needed so users can pick only the ones required in order not to sacrifice performance.

I would really appreciate your help!

[azmyrajab]

Hi @jackaixin - glad the library has been helpful for you. Had a go at this, supporting a majority of what you have requested above (and made it easy to add in more items, if you ever need to and are up for contributing)

See: https://github.com/azmyrajab/polars_ols/issues/20#issuecomment-2308885533

• Use mode="statistics" and you will get R2, t-values, p-values, etc.; it should be performant (tried to implement residual statistics with one loop and feature statistics uses faer for the xtx inversion)
• It's tested against statsmodels for plain WLS, and I supported ridge as well in case p >> n

[jackaixin]

Thanks a lot @azmyrajab for the quick turnaround. I probably should pick up rust and contribute more!