ING’s Probatus: SHAP interpretation, data similarity, and feature elimination in one Python package

ING Bank’s Probatus is a Python package for validating regression and multiclass classification models, along with the data used to develop them, according to the project’s README. It installs via pip install probatus and lists support for Python 3.10 through 3.13. The README groups the work into three modules: probatus.interpret — SHAP-based model interpretation tools. probatus.sample_similarity — comparison of two datasets using resemblance modelling, with train versus out-of-time test data given as the example. probatus.feature_elimination.ShapRFECV — cross-validated recursive feature elimination using SHAP feature importance. What the source does not show The material provided does not include benchmark results, runtime measurements, or comparisons against non-SHAP baselines. A linked ING blog post describes ShapRFECV as improved feature selection powered by SHAP, but the README itself does not reproduce numbers to support that assessment. Documentation is hosted at ing-bank.github.io/probatus, and the README links to two ING blog posts, one on ShapRFECV and one on choosing model explainability tools. That gap matters for