Reproduce “Exploring the Whole Rashomon Set of Sparse Decision Trees” 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

Rui Xin, Chudi Zhong, Zhi Chen, Takuya Takagi, Margo Seltzer, Cynthia Rudin

The paper introduces TreeFARMS, a method that fully enumerates the Rashomon set for sparse decision trees on real-world datasets. This approach enables the discovery of all almost-optimal models, demonstrating that numerous structurally diverse trees can achieve similar predictive performance, and facilitates the exact computation of Model Class Reliance.

Key insights

  • The Rashomon set for sparse decision trees contains a large number of diverse models, even for small datasets, challenging the intuition that near-optimal models are structurally similar.
  • Slight sacrifices in empirical risk can yield models with vastly different and often more desirable properties, such as enhanced interpretability or the use of specific variables.
  • Exact computation of Model Class Reliance (MCR) is possible for non-linear, discrete function classes like sparse decision trees, providing a robust measure of variable importance across the entire set of well-performing models.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.