Use of numpy dtypes as aliases for std lib types (e.g. np.int instead of int) was deprecated in numpy 1.24. This numpy version is now being installed due to the recent scikit-learn release. We don't appear to use np.int etc in alibi, but it is currently used in shap, causing CI failures. Issue tracking the shap fix: shap/shap#3063

Looks like a shap patch is incoming. Otherwise, I'm not sure of a solution except for pinning to numpy < 1.24 on our end.

Assuming shap 0.42.0 is released to fix this, we will need to bump to that and release a patch.