TrIM: Transformed Iterative Mondrian Forests for Gradient-based Dimension Reduction and High-Dimensional Regression

Ricardo Baptista, Eliza O’Reilly, and Yangxinyu Xie

Introduction

This is the official implementation of the TrIM algorithm, as described in the paper: TrIM: Transformed Iterative Mondrian Forests for Gradient-based Dimension Reduction and High-Dimensional Regression.

Requirements

• Python 3.11
• JAX

To install JAX, please follow the instructions on the JAX website

To install the remaining required packages, run:

pip install -r requirements.txt

Implementations of TrIM

The TrIM algorithm is implemented in the src/Mondrian_RF folder. Part of the code is based on the Mondrian Forests implementation by Matej Balog.

Experiments

The experiments in the paper can be reproduced by running the following scripts:

• src/Simulations.ipynb: Simulation experiments
• src/SimulationsAligned.ipynb: Simulation experiments for Weighted Mondrian Estimators
• src/Ebola.ipynb: Ebola experiments
• src/eval.py: Real data experiments on machine learning datasets