{"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/doubly-robust-policy-evaluation-and","title":"Doubly Robust Policy Evaluation and Optimization","arxiv_id":"1503.02834","date":"2015-03-10","proceeding":null,"authors":["Miroslav Dudík","Dumitru Erhan","John Langford","Lihong Li"],"abstract":"We study sequential decision making in environments where rewards are only\npartially observed, but can be modeled as a function of observed contexts and\nthe chosen action by the decision maker. This setting, known as contextual\nbandits, encompasses a wide variety of applications such as health care,\ncontent recommendation and Internet advertising. A central task is evaluation\nof a new policy given historic data consisting of contexts, actions and\nreceived rewards. The key challenge is that the past data typically does not\nfaithfully represent proportions of actions taken by a new policy. Previous\napproaches rely either on models of rewards or models of the past policy. The\nformer are plagued by a large bias whereas the latter have a large variance. In\nthis work, we leverage the strengths and overcome the weaknesses of the two\napproaches by applying the doubly robust estimation technique to the problems\nof policy evaluation and optimization. We prove that this approach yields\naccurate value estimates when we have either a good (but not necessarily\nconsistent) model of rewards or a good (but not necessarily consistent) model\nof past policy. Extensive empirical comparison demonstrates that the doubly\nrobust estimation uniformly improves over existing techniques, achieving both\nlower variance in value estimation and better policies. As such, we expect the\ndoubly robust approach to become common practice in policy evaluation and\noptimization.","url_abs":"http://arxiv.org/abs/1503.02834v1","url_pdf":"http://arxiv.org/pdf/1503.02834v1.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":"doubly-robust-policy-evaluation-and","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":"decision-making","task_name":"Decision Making"},{"task_slug":"multi-armed-bandits","task_name":"Multi-Armed Bandits"},{"task_slug":"sequential-decision-making","task_name":"Sequential Decision Making"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1503.02834","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}