{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/more-robust-doubly-robust-off-policy","title":"More Robust Doubly Robust Off-policy Evaluation","arxiv_id":"1802.03493","date":"2018-02-10","proceeding":"ICML 2018 7","authors":["Mehrdad Farajtabar","Yin-Lam Chow","Mohammad Ghavamzadeh"],"abstract":"We study the problem of off-policy evaluation (OPE) in reinforcement learning\n(RL), where the goal is to estimate the performance of a policy from the data\ngenerated by another policy(ies). In particular, we focus on the doubly robust\n(DR) estimators that consist of an importance sampling (IS) component and a\nperformance model, and utilize the low (or zero) bias of IS and low variance of\nthe model at the same time. Although the accuracy of the model has a huge\nimpact on the overall performance of DR, most of the work on using the DR\nestimators in OPE has been focused on improving the IS part, and not much on\nhow to learn the model. In this paper, we propose alternative DR estimators,\ncalled more robust doubly robust (MRDR), that learn the model parameter by\nminimizing the variance of the DR estimator. We first present a formulation for\nlearning the DR model in RL. We then derive formulas for the variance of the DR\nestimator in both contextual bandits and RL, such that their gradients\nw.r.t.~the model parameters can be estimated from the samples, and propose\nmethods to efficiently minimize the variance. We prove that the MRDR estimators\nare strongly consistent and asymptotically optimal. Finally, we evaluate MRDR\nin bandits and RL benchmark problems, and compare its performance with the\nexisting methods.","url_abs":"http://arxiv.org/abs/1802.03493v2","url_pdf":"http://arxiv.org/pdf/1802.03493v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"more-robust-doubly-robust-off-policy","repo_url":"https://github.com/PlaytikaOSS/pybandits","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"multi-armed-bandits","task_name":"Multi-Armed Bandits"},{"task_slug":"off-policy-evaluation","task_name":"Off-policy evaluation"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1802.03493","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}