Add methods for performing Wald-tests

[MartinStancsicsQC]

Checklist

• Added a CHANGELOG.rst entry

Summary

This PR adds some support for doing inference on estimated GLMs. Wald-tests can be performed based on a restriction matrix and corresponding RHS vector (lower-level, flexible), a list of feature names or a list of term names.

Notes

• This branch is based on the store-cov-matrix branch. Therefore, PR Add the option to store the covariance matrix to avoid recomputing it #661 should be merged before this one. I am opening this draft PR to get some feedback until that happens.
• Term names is not a concept in glum yet. The idea is based on Tabmat PR #278 and the closed Glum PR Create feature names for non-pandas input, too #655. Either these will have to be merged first, or the GeneralizedLinearRegressorBase.wald_test_term_name method will have to be removed from this PR until term names are implemented.

[MatthiasSchmidtblaicherQC]

Looks good to me. I have some small stylistic points and a suggestion for a check that we could do.

[MarcAntoineSchmidtQC]

Looks really good.
Please look at my suggestion for renaming the wald test methods. Otherwise, I think this can be merged.
🎉

[MartinStancsicsQC]

Thank you @MarcAntoineSchmidtQC! The methods have been renamed. Should it go into main or glum-v3?

[MartinStancsicsQC]

Note: windows conda-build also fails on main. (was fine in the morning...)

[MartinStancsicsQC]

Related mamba issue: mamba-org/mamba#2770

[jonashaag]

Are you pinning Mamba to 1.4.2 on purpose?

[MartinStancsicsQC]

I don't think so. I'm not even sure where that pin comes from.

[MartinStancsicsQC]

@jonashaag I don't think it's about the libmamba version. The CI for the same commit was green this morning but red when I rerun it this afternoon.. Based on the logs, the difference between the build environment is that the passing one uses libsolv 0.7.24 build 1, while the failing one uses libsolv 0.7.24 build 2. (there is also a difference in the patch version of wheel but I don't think that's the issue.) Is it possible that it is a libsolv problem?

[jonashaag]

@pavelzw it looks like you made the libsolv change, can you have a look please?

[MartinStancsicsQC]

Never mind, libsolv build 2 was indeed broken and build 3 seem to have solved our issues