Straightforward implementations of interpretable ML models + demos of how to use various interpretability techniques. Code is optimized for readability. Pull requests very welcome!

Docs • Implementations of imodels • Demo notebooks

Scikit-learn style wrappers/implementations of different interpretable models. The interpretable models can be easily installed and used:

pip install git+https://github.com/csinva/imodels (see here for more help)

from imodels import RuleListClassifier, GreedyRuleListClassifier, SkopeRulesClassifier, IRFClassifier
from imodels import SLIMRegressor, RuleFitRegressor

model = RuleListClassifier()  # initialize Bayesian Rule List
model.fit(X_train, y_train)   # fit model
preds = model.predict(X_test) # discrete predictions: shape is (n_test, 1)
preds_proba = model.predict_proba(X_test) # predicted probabilities: shape is (n_test, n_classes)

• bayesian rule list (docs, ref implementation, paper) - learn a compact rule list
• rulefit (docs, ref implementation, paper) - find rules from a decision tree and build a linear model with them
• skope-rules (docs, ref implementation) - extracts rules from base estimators (e.g. decision trees) then tries to deduplicate them
• sparse integer linear model (docs, cvxpy implementation, paper)
• greedy rule list (docs, ref implementation) - uses CART to learn a list (only a single path), rather than a decision tree
• (in progress) iterative random forest (docs, ref implementation, paper)
• (in progress) optimal classification tree (docs, ref implementation, paper) - learns succinct trees using global optimization rather than greedy heuristics
• (coming soon) rule ensembles - e.g. SLIPPER, Lightweight Rule Induction, MLRules
• (coming soon) gams
• (coming soon) symbolic regression

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

• model_based.ipynb - how to use different interpretable models and examples with the imodels package
  • see an example of using this package for deriving a clinical decision rule in this nb

After fitting models, we can also do posthoc analysis as shown in these two notebooks:

• posthoc.ipynb - different simple analyses to interpret a trained model
• uncertainty.ipynb - code to get uncertainty estimates for a model

• Readings
  • Interpretable Machine Learning: A Guide for Making Black Box Models Explainable (molnar 2019, pdf) - book on interpretable ML
  • Interpretable machine learning: definitions, methods, and applications (murdoch et al. 2019, pdf) - good quick review on interpretable ML
  • Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead (rudin 2019, pdf) - good explanation of why one should use interpretable models
  • Review on evaluating interpretability (doshi-velez & kim 2017, pdf)
• Reference implementations (also linked above): the code here heavily derives from (and in some case is just a wrapper for) the wonderful work of previous projects. We seek to to extract out, combine, and maintain select relevant parts of these projects.
  • sklearn-expertsys - by @tmadl and others based on original code by Ben Letham
  • rulefit - by @christophM
  • skope-rules - by the skope-rules team

For updates, star the repo, see this related repo, or follow @chandan_singh96. Please make sure to give authors of original methods / base implementations appropriate credit!