A package for tree-based statistical estimation and inference using optimal decision trees. ODTlearn provides implementations of StrongTrees [1], FairTrees [2], RobustTrees [3] for classification, and Prescriptive Trees [4] for prescription.

The package documentation contains usage examples and method reference.

The latest stable version can be installed from PyPI with the command:

pip install odtlearn

The current development version can be installed from source with the following command:

pip install git+https://github.com/D3M-Research-Group/odtlearn.git#egg=odtlearn

To use Gurobi with ODTlearn, you must have a valid Gurobi License. Free licenses are available for academic use and additional methods for obtaining a Gurobi license can be found here.

Python-MIP provides CBC binaries for 64-bit versions of Windows, Linux, and MacOS that run on Intel hardware, however we have observed that these binaries do not seem to work properly with lazy constraint generation, which is used in some of our MIO formulations. Thus, to ensure expected behavior when using ODTlearn, we strongly recommend building CBC from source. Below are the steps needed to compile CBC from source using coinbrew.

mkdir CBC
cd CBC
wget -nH https://raw.githubusercontent.com/coin-or/coinbrew/master/coinbrew
chmod u+x coinbrew 
bash coinbrew fetch Cbc@master --no-prompt
bash coinbrew build Cbc@stable/2.10

export DYLD_LIBRARY_PATH=/PATH/TO/CBC/dist/lib
export PMIP_CBC_LIBRARY=/PATH/TO/CBC/dist/lib/PLATFORM_SPECIFIC_SHARED_LIB

The last two steps are critical for ensuring that ODTlearn (through Python-MIP) uses the correct CBC binary. For Windows and MacOS the shared library name is libCbc.dll and libCbc.dylib, respectively. For Linux, the shared library name is libCbc.so. To ensure that the environment variables persist, we suggest adding the last two lines to your .zshrc or .bashrc file.

This project uses black to format code and flake8 for linting. We also support pre-commit to ensure these have been run. To configure your local environment please install these development dependencies and set up the commit hooks.

pip install black flake8 pre-commit
pre-commit install

• [1] Aghaei, S., Gómez, A., & Vayanos, P. (2021). Strong optimal classification trees. arXiv preprint arXiv:2103.15965. https://arxiv.org/abs/2103.15965.
• [2] Jo, N., Aghaei, S., Benson, J., Gómez, A., & Vayanos, P. (2022). Learning optimal fair classification trees. arXiv preprint arXiv:2201.09932. https://arxiv.org/pdf/2201.09932.pdf
• [3] Justin, N., Aghaei, S., Gomez, A., & Vayanos, P. (2021). Optimal Robust Classification Trees. In The AAAI-22 Workshop on Adversarial Machine Learning and Beyond. https://openreview.net/pdf?id=HbasA9ysA3
• [4] Jo, N., Aghaei, S., Gómez, A., & Vayanos, P. (2021). Learning optimal prescriptive trees from observational data. arXiv preprint arXiv:2108.13628. https://arxiv.org/pdf/2108.13628.pdf