Machine Learning Interpretability

Learn how to interpret and explain machine-learning models — from intrinsically interpretable models like linear regression and decision trees, to post-hoc methods like permutation importance, partial dependence, ALE, LIME and SHAP.

This repository contains the practical notebooks for the Machine Learning Interpretability course. The examples cover both intrinsically explainable models and model-agnostic post-hoc explainability methods, using scikit-learn, statsmodels, XGBoost, LightGBM, eli5, LIME, and SHAP.

Course launch: November 2023

Last updated: September 2026

Status: Actively maintained

What you will learn

• Tell apart local vs. global, and intrinsic vs. post-hoc interpretability.
• Interpret linear and logistic regression, decision trees, random forests, and gradient boosting machines natively.
• Quantify global feature importance with permutation importance, partial dependence plots, and accumulated local effects (ALE).
• Explain individual predictions with individual conditional expectation (ICE) plots, surrogate models, LIME, and SHAP.
• Compare exact and approximate Shapley value calculations, including TreeSHAP, the exact explainer, and the permutation explainer.
• Use eli5, LIME, and SHAP in practice, and understand the assumptions and trade-offs behind each method.

Course contents

1. Machine Learning Interpretability

   • Interpretability in the context of Machine Learning
   • Local vs Global Interpretability
   • Intrinsically explainable models
   • Post-hoc explainability methods
   • Challenges to interpretability
   • How to make models more explainable

2. Intrinsically Explainable Models

   • Linear regression
   • Logistic regression
   • Decision trees
   • Random forests
   • Gradient boosting machines

3. Post-hoc methods - Global explainability

   • Permutation feature importance
   • Partial dependence plots
   • Accumulated local effects (ALE)

4. Post-hoc methods - Local explainability

   • Individual conditional expectation (ICE)
   • Surrogate models
   • LIME
   • SHAP

5. Featuring the following Python interpretability libraries

   • scikit-learn
   • statsmodels
   • eli5
   • LIME
   • SHAP

Getting started

Clone the repository and set up a dedicated Python environment (Python 3.11+).

Take a look at requirements.txt and install what you need for the sections you plan to work through. keras and tensorflow are listed separately as optional — they're only needed for the LIME text and image notebooks in 14-lime, so there's no need to install them if you're not running those.

Then start Jupyter and open the notebooks in numerical order.

Course

For lectures, explanations, and the complete learning path, visit the online course.