API Reference / algorithms /

glassbox/

Interpretable-by-design ("glassbox") model family.

Glassbox models are accurate enough to use in production yet simple enough for a human to read the entire model from top to bottom. Unlike post-hoc explanations (SHAP, LIME) that approximate a black box, every prediction these models make is the exact sum of the parts you can inspect.

Algorithms

  • ExplainableBoostingClassifier / ExplainableBoostingRegressor: Additive
models of per-feature shape functions (EBM / GA2M). Each feature's bin-to-score mapping is learned and exposed, so the whole model reads as intercept + f_0(x_0) + f_1(x_1) + ....
  • RuleFitClassifier / RuleFitRegressor: A forest is distilled into a
small set of human-readable feature <= t / feature > t rules, then a sparse linear model is fit over the rules plus the original features.

See Also

trees

The tree learners RuleFit distils.
linear

Plain (non-additive) linear baselines.

Modules

ebm

Explainable Boosting Machine (EBM / GA2M): additive glassbox models....

2 classes · 0 functions
rulefit

RuleFit: a forest distilled into a sparse linear model of readable rules....

2 classes · 0 functions