Doubly Robust Off-policy Value Evaluation for Reinforcement Learning

TL;DR

This paper introduces a doubly robust estimator for off-policy evaluation in reinforcement learning that is unbiased, has low variance, and is effective in benchmark tests, advancing safe policy improvement methods.

Contribution

It extends the doubly robust estimator to sequential decision-making, providing a method with guaranteed unbiasedness and reduced variance, matching theoretical lower bounds in some cases.

Findings

The estimator is unbiased and has lower variance than importance sampling.

It performs accurately on benchmark problems.

It can be used effectively for safe policy improvement.

Abstract

We study the problem of off-policy value evaluation in reinforcement learning (RL), where one aims to estimate the value of a new policy based on data collected by a different policy. This problem is often a critical step when applying RL in real-world problems. Despite its importance, existing general methods either have uncontrolled bias or suffer high variance. In this work, we extend the doubly robust estimator for bandits to sequential decision-making problems, which gets the best of both worlds: it is guaranteed to be unbiased and can have a much lower variance than the popular importance sampling estimators. We demonstrate the estimator's accuracy in several benchmark problems, and illustrate its use as a subroutine in safe policy improvement. We also provide theoretical results on the hardness of the problem, and show that our estimator can match the lower bound in certain scenarios.