Sample Complexity and Decision-Theoretic Guarantees for Bayesian Model Averaging over Decision Trees with Catalan-Exponential Priors
Title: Rational Commitment Thresholds and Sample Complexity in Bayesian Model Averaging for Decision Trees with Catalan-Exponential Priors
Abstract: This study investigates the conditions under which Bayesian model averaging (BMA) weights applied to decision trees contain adequate epistemic information to warrant committed exploitation of the averaged distribution. By focusing on Bayesian decision trees (BDTs) that employ Dirichlet-Multinomial leaf models alongside a Catalan-exponential prior on tree size (Schetinin & Jakaite, 2025), we provide a closed-form answer to this inquiry. Consequently, we establish a comprehensive non-asymptotic theoretical framework governing rational commitment thresholds.
Source: arXiv Generated at: 2026-06-02 00:00:00 UTC
