Asymptotically Efficient Off-Policy Evaluation for Tabular Reinforcement Learning

01/29/2020
by   Ming Yin, et al.
12

We consider the problem of off-policy evaluation for reinforcement learning, where the goal is to estimate the expected reward of a target policy π using offline data collected by running a logging policy μ. Standard importance-sampling based approaches for this problem suffer from a variance that scales exponentially with time horizon H, which motivates a splurge of recent interest in alternatives that break the "Curse of Horizon" (Liu et al. 2018, Xie et al. 2019). In particular, it was shown that a marginalized importance sampling (MIS) approach can be used to achieve an estimation error of order O(H^3/ n) in mean square error (MSE) under an episodic Markov Decision Process model with finite states and potentially infinite actions. The MSE bound however is still a factor of H away from a Cramer-Rao lower bound of order Ω(H^2/n). In this paper, we prove that with a simple modification to the MIS estimator, we can asymptotically attain the Cramer-Rao lower bound, provided that the action space is finite. We also provide a general method for constructing MIS estimators with high-probability error bounds.

READ FULL TEXT

page 1

page 2

page 3

page 4

research
12/07/2022

Low Variance Off-policy Evaluation with State-based Importance Sampling

In off-policy reinforcement learning, a behaviour policy performs explor...
research
04/11/2023

A Tale of Sampling and Estimation in Discounted Reinforcement Learning

The most relevant problems in discounted reinforcement learning involve ...
research
01/09/2023

Minimax Weight Learning for Absorbing MDPs

Reinforcement learning policy evaluation problems are often modeled as f...
research
06/08/2019

Optimal Off-Policy Evaluation for Reinforcement Learning with Marginalized Importance Sampling

Motivated by the many real-world applications of reinforcement learning ...
research
12/14/2022

Scaling Marginalized Importance Sampling to High-Dimensional State-Spaces via State Abstraction

We consider the problem of off-policy evaluation (OPE) in reinforcement ...
research
07/07/2020

Near Optimal Provable Uniform Convergence in Off-Policy Evaluation for Reinforcement Learning

The Off-Policy Evaluation aims at estimating the performance of target p...
research
05/16/2022

Efficient Algorithms for Planning with Participation Constraints

We consider the problem of planning with participation constraints intro...

Please sign up or login with your details

Forgot password? Click here to reset