Estimating the Maximum Expected Value: An Analysis of (Nested) Cross Validation and the Maximum Sample Average

Abstract

We investigate the accuracy of the two most common estimators for the maximum expected value of a general set of random variables: a generalization of the maximum sample average, and cross validation. No unbiased estimator exists and we show that it is non-trivial to select a good estimator without knowledge about the distributions of the random variables. We investigate and bound the bias and variance of the aforementioned estimators and prove consistency. The variance of cross validation can be significantly reduced, but not without risking a large bias. The bias and variance of different variants of cross validation are shown to be very problem-dependent, and a wrong choice can lead to very inaccurate estimates.

Cited by in corpus (5)

• Controlling Overestimation Bias with Truncated Mixture of Continuous Distributional Quantile Critics
• Regularized Softmax Deep Multi-Agent $Q$-Learning
• Action Candidate Based Clipped Double Q-learning for Discrete and Continuous Action Tasks
• Deep Reinforcement Learning with Weighted Q-Learning
• Cross Learning in Deep Q-Networks