An interpretable DT ensemble method based on the Mixture of Experts architecture. The implementation is inspired by a similar approach called Mixture of Expert Trees: https://arxiv.org/abs/1906.06717 .
The main class can be found in modt/modt.py. For a simplistic application execute run_modt.py, more advanced notebooks can be found in the examples folder. Installation instructions are given below.
The Expectation-Maximization (EM) training algorithm iteratively optimizes a set of decision trees and associated regions:
Decision area of the final gating function:
Resulting decision trees:
DT0 red region | DT1 green region | DT2 blue region |
---|---|---|
MoDT is used with the steel plates faults dataset: https://archive.ics.uci.edu/ml/datasets/Steel+Plates+Faults
The DT visualizations are created with dtreeviz: https://github.com/parrt/dtreeviz
Minimally, to run run_modt.py, the packages in requirments_minimal.txt must be installed.
pip install -r requirements_minimal.txt
For optional functionalities, and to run the example notebooks, use requirements_maximal.txt.
pip install -r requirements_maximal.txt
The packages dtreeviz and celluloid require additional codices. Optimal Decision Trees is a commercial software from Interpretable AI.