MAIF/shapash

🔅 Shapash: User-friendly Explainability and Interpretability to Develop Reliable and Transparent Machine Learning Models

What it solves

Shapash is designed to make machine learning models more interpretable and comprehensible for both data scientists and non-experts. It bridges the gap between complex model outputs and human-readable explanations by providing clear visualizations and explicit labels, reducing the need for technical jargon when sharing results.

How it works

It acts as an overlay package that utilizes interpretability backends like SHAP or LIME to compute feature contributions. It then transforms these technical contributions into a user-friendly interface, including a Webapp for navigating between global and local explainability, and a standalone HTML report for auditing purposes. It supports a wide range of models (e.g., Catboost, XGBoost, LightGBM, Scikit-Learn) and integrates with common preprocessing tools like ColumnTransformers and category-encoders.

Who it’s for

  • Data Scientists: To quickly understand, audit, and validate their models.
  • Non-Data Experts: To receive clear, summarized explanations of model predictions without needing deep technical knowledge.
  • Audit and Compliance Teams: To review model behavior through comprehensive HTML reports for AI governance.

Highlights

  • Interactive Webapp: Allows seamless navigation between global feature importance and local prediction explanations.
  • Audit Reports: Generates standalone HTML reports to serve as a basis for model auditing.
  • Model Agnostic: Compatible with numerous popular ML models and supports custom contribution backends.
  • Production Ready: Includes a SmartPredictor object for deploying summarized local explanations via API or batch mode.
  • Explainability Metrics: Provides tools to evaluate the quality of explanations using stability, consistency, and compacity metrics.

Related

  • Project
    microsoft/responsible-ai-toolbox

    A suite of model and data exploration tools that enable developers to identify errors, assess fairness, and interpret AI model predictions for more responsible deployment.

  • Project
    deel-ai/xplique

    A Python toolkit for Explainable AI (XAI) that provides tools for attribution, feature visualization, and concept extraction to help users understand complex neural network models.

  • Project
    interpretml/interpret

    An open-source framework that provides both inherently interpretable glassbox models and tools to explain complex blackbox machine learning systems.

  • Project
    csinva/imodels

    Python library offering a large suite of scikit‑learn‑compatible, interpretable models (rule sets, rule lists, trees, sparse linear models, etc.) with utilities for post‑hoc distillation and AutoML.

  • Project
    PAIR-code/lit

    A visual, interactive ML model-understanding tool that supports text, image, and tabular data to help developers debug and interpret model predictions.