Dynamic Sampling Allocation for Selecting a Good Enough Alternative

Gongbo Zhang, Chun-Hung Chen, Qing-Shan Jia, Yijie Peng

Abstract: We consider the problem of selecting a good enough alternative from a finite set of alternatives. Instead of selecting the exactly best alternative, our work aims to maximize the probability of correctly selecting an alternative in an acceptable subset. Under a Bayesian framework, we formulate the problem as a stochastic control problem. We propose a dynamic allocation scheme for selecting a good enough alternative, which optimizes a value function approximation one-step ahead. Numerical results demonstrate the proposed sampling procedure is more efficient than other sampling allocation methods.