gverbock/probatus

Validation of binary classifiers and data used to develop them

probatus

Overview

Probatus is a python package that helps validate binary classification models and the data used to develop them. Main features:

• probatus.interpret provides shap-based model interpretation tools
• probatus.metric_volatility provides tools using bootstrapping and/or different random seeds to assess metric volatility/stability.
• probatus.sample_similarity to compare two datasets using resemblance modelling, f.e. train with out-of-time test.
• probatus.feature_elimination.ShapRFECV provides cross-validated Recursive Feature Elimination using shap feature importance.
• probatus.missing_values compares performance gains of different missing values imputation strategies for a given model.

Installation

pip install probatus

Documentation

Documentation at ing-bank.github.io/probatus/.

You can also check out blog posts about Probatus:

• Open-sourcing ShapRFECV — Improved feature selection powered by SHAP.
• Model Explainability — How to choose the right tool?

Contributing

To learn more about making a contribution to probatus, please see CONTRIBUTING.md.