State Relevance for Off-Policy Evaluation

TL;DR

This paper introduces OSIRIS, a new importance sampling estimator for off-policy evaluation that reduces variance by omitting certain state likelihood ratios, maintaining unbiasedness under specific conditions.

Contribution

The paper proposes OSIRIS, a novel variance reduction technique for importance sampling in off-policy evaluation, with formal conditions for unbiasedness and empirical validation.

Findings

OSIRIS achieves lower variance than traditional importance sampling.

OSIRIS remains unbiased under specified conditions.

Empirical results confirm variance reduction benefits.

Abstract

Importance sampling-based estimators for off-policy evaluation (OPE) are valued for their simplicity, unbiasedness, and reliance on relatively few assumptions. However, the variance of these estimators is often high, especially when trajectories are of different lengths. In this work, we introduce Omitting-States-Irrelevant-to-Return Importance Sampling (OSIRIS), an estimator which reduces variance by strategically omitting likelihood ratios associated with certain states. We formalize the conditions under which OSIRIS is unbiased and has lower variance than ordinary importance sampling, and we demonstrate these properties empirically.