This repository contains notebook code and data for the paper: Explainable AI for Trees: From Local Explanations to Global Understanding
This paper presents the first exact game theoretic solution for explaining predictions from tree ensemble models, the most widely used non-linear machine learning models; this solution is the only one that guarantees desirable properties and so enables new insights into individual predictions and global behavior of the model.
Ways to explore the results and code behind the paper:
-
Use TreeExplainer to explain your own tree-based models with the popular Python shap package.
- System requirements: Any recent Mac, Windows, or Linux OS. See https://anaconda.org/conda-forge/shap/files for a list of systems with precompiled binaries the codes is unit tested on.
- Installation: See https://github.com/slundberg/shap (install time is typically < 1 min)
- Demo: See the TreeExplainer demos at https://github.com/slundberg/shap, the demo on the front page runs in a few seconds on any modern computer.
-
Check out the local explanation benchmark results.
-
Download the NHANES I data (the chronic kidney disease and hospital datasets require approval for access)
-
Browse the released notebooks (still under construction).