imodels

As interpretability has become a requirement in high-stakes domains like healthcare, credit scoring, and criminal justice, the ability to use models whose decision logic is fully transparent (not just explainable post-hoc) has become both a regulatory expectation and an ethical consideration.

imodels implements techniques like rule lists, rule sets, symbolic regression, and optimal sparse decision trees.

The package includes implementations of algorithms that produce genuinely interpretable models: Bayesian Rule Lists, which learn probabilistic if-then rule sequences; FIGS (Fast Interpretable Greedy-Tree Sums), which fit additive tree models that remain comprehensible; and Greedy Rule Lists that balance predictive performance against rule complexity.

These models can often match or approach the accuracy of more complex models on tabular data while producing decision logic that domain experts can review, validate, and critique based on their knowledge of the problem.

Clinical decision support developers who need physicians to understand and trust model recommendations, credit risk teams whose models must satisfy explainability requirements under financial regulation, and researchers studying the accuracy-interpretability frontier in machine learning use imodels to access state-of-the-art interpretable modeling algorithms without implementing them from scratch.

The package's scikit-learn compatible interface means interpretable models can be evaluated directly alongside black-box competitors, enabling rigorous comparison of the accuracy cost of interpretability constraints on specific datasets.