Explainerdashboard
The AI research community has always considered interactive visualization to be an important approach for interpreting ML model predictions. In this section, we will cover Explainerdashboard, which is an interesting Python framework that can spin up a comprehensive interactive dashboard covering various aspects of model explainability with just minimal lines of code. Although this framework supports only scikit-learn-compatible models (including XGBoost, CatBoost, and LightGBM), it can provide model-agnostic global and local explainability. Currently, it supports SHAP-based feature importance and interactions, PDPs, model performance analysis, what-if model analysis, and even decision-tree-based breakdown analysis plots.
The framework allows customization of the dashboard, but I think the default version includes all supported aspects of model explainability. The generated web-app-based dashboards can be exported as static web pages directly from a live dashboard...