Scaled Rate Optimization for Beta-Binomial Models

Abstract

Rates of binomial processes are modeled using beta-binomial distributions (for example, from Beta Regression). We treat the offline optimization scenario and then the online one, where we optimize the exploration-exploitation problem. The rates given by two processes are compared through their distributions, but we would like to optimize the net payout (given a constant value per successful event, unique for each of the processes). The result is an analytically-closed, probabilistic, hypergeometric expression for comparing the payout distributions of two processes. To conclude, we contrast this Bayesian result with an alternative frequentist approach and find 4.5 orders of magnitude improvement in performance, for a numerical accuracy level of 0.01%.