Efficient Evaluation of Natural Stochastic Policies in Offline Reinforcement Learning
This work addresses a specific challenge in offline RL for practitioners by improving policy evaluation methods to handle natural stochastic policies, which can enhance implementability and mitigate overlap issues, though it is incremental in nature.
The paper tackles the problem of efficiently evaluating natural stochastic policies in offline reinforcement learning, which are defined relative to a behavior policy rather than explicitly specified, and it derives efficiency bounds for two policy types and proposes nonparametric estimators that achieve these bounds under lax conditions.
We study the efficient off-policy evaluation of natural stochastic policies, which are defined in terms of deviations from the behavior policy. This is a departure from the literature on off-policy evaluation where most work consider the evaluation of explicitly specified policies. Crucially, offline reinforcement learning with natural stochastic policies can help alleviate issues of weak overlap, lead to policies that build upon current practice, and improve policies' implementability in practice. Compared with the classic case of a pre-specified evaluation policy, when evaluating natural stochastic policies, the efficiency bound, which measures the best-achievable estimation error, is inflated since the evaluation policy itself is unknown. In this paper, we derive the efficiency bounds of two major types of natural stochastic policies: tilting policies and modified treatment policies. We then propose efficient nonparametric estimators that attain the efficiency bounds under very lax conditions. These also enjoy a (partial) double robustness property.