csinva/imodels

Interpretable ML package 🔍 for concise, transparent, and accurate predictive modeling (sklearn-compatible).

What it solves

Modern machine learning models are often complex "black-box" systems that are difficult to interpret. imodels provides a collection of state-of-the-art interpretable models that can replace complex models (like random forests) with simpler, transparent alternatives (like rule lists or small trees) without sacrificing predictive accuracy.

How it works

The package implements a variety of interpretable modeling techniques that follow the scikit-learn estimator API (fit and predict methods). It offers four primary forms of interpretable output:

  • Rule sets: Sparse linear models or combined rules extracted from trees.
  • Rule lists: Sequential lists of rules (e.g., Bayesian or Greedy rule lists).
  • Rule trees: Decision trees using various fitting methods (e.g., CART, C4.5, TAO).
  • Algebraic models: Sparse integer linear models or Generalized Additive Models (GAMs).

It also includes wrappers for distillation (converting a black-box model into an interpretable one) and AutoML for automatic model selection.

Who it’s for

Data scientists and researchers who need predictive models that are transparent, concise, and easy to explain to stakeholders, particularly in high-stakes fields like clinical decision-making.

Highlights

  • Scikit-learn compatible: Works seamlessly with sklearn pipelines and grid searches.
  • Diverse model library: Includes everything from Bayesian rule lists to Gaussian Process GAMs.
  • Model distillation: Ability to train a complex model and then distill it into a simpler, interpretable version.
  • AutoML integration: Built-in tools to automatically fit and select the best interpretable model.

Related

  • Project
    jpmml/sklearn2pmml

    A Python package that converts Scikit-Learn pipelines to PMML format to improve model portability and deployment across different environments.

  • Project
    interpretml/interpret

    InterpretML is an open‑source Python library that combines interpretable glass‑box models (like Explainable Boosting Machines) and post‑hoc explainers (SHAP, LIME, etc.) with interactive visual dashboards, supporting privacy‑preserving training and large‑scale workloads.

  • Project
    feature-engine/feature_engine

    A Python library providing Scikit-learn compatible transformers for comprehensive feature engineering and selection to improve machine learning model performance.

  • Project
    Nixtla/mlforecast

    A scalable framework for time series forecasting that allows any scikit-learn compatible machine learning model to be used for forecasting millions of series.

  • Project
    MAIF/shapash

    A Python library that makes machine learning models interpretable for everyone by providing user-friendly visualizations, an interactive web app, and audit reports.