This is a fork of dl8.5, which adds lower-bound guessing, as outlined in the paper Fast Sparse Decision Tree Optimization via Reference Ensembles.

Installation

python3 setup.py install

Note: you may need to manually uninstall dl8.5 with pip if you have used the pip installation of dl8.5 before.

Differences from dl8.5

Lower-bound guessing is handled via an extra input warm, which would be a vector of the predictions of a reference model for each point in the training set. (label 0 = reference model's prediction was incorrect for this point, 1 = reference model's prediction was correct for this point)

All use cases for dl8.5 from prior to September 2021 are supported, although we currently do not support the use of a reference model with the hyperparameter iterative=true.

Relevant Papers

• Aglin, G., Nijssen, S., Schaus, P. Learning optimal decision trees using caching branch-and-bound search. In AAAI. 2020.
• McTavish, H., Zhong, C., Achermann, R., Karimalis, I., Chen, J., Rudin, C., & Seltzer, M. Fast Sparse Decision Tree Optimization via Reference Ensembles. In AAAI. 2022.

Note: This fork was created prior to dl8.5's September 2021 updates to support non-classification tasks.