Interpretability demos + implementations

Demos of how to use various interpretability techniques (accompanying slides here) and code for implementations of interpretable machine learning models.

Implementations of interpretable models

Provides scikit-learn style wrappers/implementations of different interpretable models (see readmes in individual folders within imodels for details)

• bayesian rule list (based on this implementation)
• rulefit (based on this implementation)
• sparse integer linear model (simple implementation based on cvxpy)
• optimal classification tree (based on this implementation)
• greedy rule list (based on this impelementation) - grows a CART tree that only goes down one path, resulting in a list rather than a tree

The interpretable models within the imodels folder can be easily installed and used.

pip install git+https://github.com/csinva/interpretability-implementations-demos

from imodels import RuleListClassifier, RuleFit, GreedyRuleList, SLIM
model = RuleListClassifier() # Bayesian Rule List
model.fit(X_train, y_train)
model.score(X_test, y_test)
preds = model.predict(X_test)

Demo notebooks

The demos are contained in 3 main notebooks, following this cheat-sheet:

1. model_based.ipynb - how to use different interpretable models
2. posthoc.ipynb - different simple analyses to interpret a trained model
3. uncertainty.ipynb - code to get uncertainty estimates for a model

References / further reading

• high-level review on interpretable machine learning
• book on interpretable machine learning
• review on black-blox explanation methods
• review on variable importance
• for updates, star the repo, see this related repo, or follow @chandan_singh96