Reproduce “bsnsing: A decision tree induction method based on recursive optimal boolean rule composition” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yanchao Liu
BSNSING introduces a decision tree induction method that uses a Mixed-Integer Programming (MIP) formulation to optimally select multivariate boolean split rules at each node, explicitly maximizing Gini reduction. The method, supported by a highly efficient custom implicit enumeration (ENUM) algorithm, demonstrates competitive predictive performance (leading in AUC for binary classification) with substantially faster training speeds compared to other optimal classification tree approaches.
Key insights
- A Mixed-Integer Programming (MIP) formulation can directly optimize the Gini reduction criterion, a non-convex measure, for decision tree splits.
- Developing a specialized implicit enumeration (ENUM) algorithm can make exact optimization of split rules computationally feasible, often outperforming commercial MIP solvers by orders of magnitude.
- Embedding optimal split rule selection within a recursive partitioning framework offers a practical balance between model performance and computational efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.