mljar/supertree

What it solves

It solves the problem of static decision tree visualizations (like those from scikit-learn) becoming unreadable as the tree depth increases. It provides an interactive way to inspect, explain, and debug complex decision trees and ensembles without node overlap or shrinking text.

How it works

supertree renders an interactive HTML-based visualization directly within Jupyter Notebooks, JupyterLab, Google Colab, or VS Code notebooks. It integrates with popular machine learning libraries to extract tree structures and presents them as a navigable interface.

Who it’s for

Data scientists and ML engineers who need to visualize and analyze decision trees, random forests, or gradient boosting models from scikit-learn, XGBoost, LightGBM, or ONNX.

Highlights

• Interactive Navigation: Zoom, pan, and collapse/expand nodes to manage complex trees.
• Broad Compatibility: Supports a wide range of classifiers and regressors from scikit-learn, LightGBM, XGBoost, and ONNX.
• Ensemble Support: Ability to switch between and visualize individual trees within a forest.
• Sample Tracing: Highlight the decision path of a single sample through the tree.
• Detailed Inspection: Full node statistics available on hover and the ability to export visualizations as SVG.