Christian Lorentzen

I would use the model predictions instead of the score per obs, pretty much a blend of `decompose` and `compute_bias`: ```python def compute_score( y_obs, y_pred, feature, weights, scoring_function, functional, level,...

I guess uncertainty / confidence intervals would be enough. As you say, for bias there is a universal reference, i.e. zero, for scores all pairwise comparison are options, that's way...

#162 implemented partial dependece. So `compute_partial_dependence` is more about making it public. The plotting needs a bit of code though. Under which module should it be placed? A new one...

@OmarManzoor @adam2392 Thanks for your reviews. Usually, within the scikit-learn project, one of the reviewers merges.

@adam2392 Congratulation to your first merge🚀

@StefanieSenger It seems you have invested quite a lot into this PR. As of version 2.0, numpy has weighted quantile and nanquantile. How about just using those?

> we also need our own function to support array api inputs (eg a pytorch input allocated on a GPU device) as nan + weight support is not part of...

> > @lorentzenchr, has @ogrisel convinced you this effort is worthwhile? What is your take on the question of naming variables? > > I think it's worthwhile, at least in...

> @lorentzenchr, is this PR ready for merge now? I'll approve when a test vs numpy's weighted quantile function is added.

Please try to make CI green. If you need help to achieve this you can ask for support.