Deprecated: $wgMWOAuthSharedUserIDs=false is deprecated, set $wgMWOAuthSharedUserIDs=true, $wgMWOAuthSharedUserSource='local' instead [Called from MediaWiki\HookContainer\HookContainer::run in /var/www/html/w/includes/HookContainer/HookContainer.php at line 135] in /var/www/html/w/includes/Debug/MWDebug.php on line 372
Doubly Robust Off-Policy Value and Gradient Estimation for Deterministic Policies - MaRDI portal

Deprecated: Use of MediaWiki\Skin\SkinTemplate::injectLegacyMenusIntoPersonalTools was deprecated in Please make sure Skin option menus contains `user-menu` (and possibly `notifications`, `user-interface-preferences`, `user-page`) 1.46. [Called from MediaWiki\Skin\SkinTemplate::getPortletsTemplateData in /var/www/html/w/includes/Skin/SkinTemplate.php at line 691] in /var/www/html/w/includes/Debug/MWDebug.php on line 372

Deprecated: Use of QuickTemplate::(get/html/text/haveData) with parameter `personal_urls` was deprecated in MediaWiki Use content_navigation instead. [Called from MediaWiki\Skin\QuickTemplate::get in /var/www/html/w/includes/Skin/QuickTemplate.php at line 131] in /var/www/html/w/includes/Debug/MWDebug.php on line 372

Doubly Robust Off-Policy Value and Gradient Estimation for Deterministic Policies

From MaRDI portal
Publication:6342251

arXiv2006.03900MaRDI QIDQ6342251

Author name not available (Why is that?)

Publication date: 6 June 2020

Abstract: Offline reinforcement learning, wherein one uses off-policy data logged by a fixed behavior policy to evaluate and learn new policies, is crucial in applications where experimentation is limited such as medicine. We study the estimation of policy value and gradient of a deterministic policy from off-policy data when actions are continuous. Targeting deterministic policies, for which action is a deterministic function of state, is crucial since optimal policies are always deterministic (up to ties). In this setting, standard importance sampling and doubly robust estimators for policy value and gradient fail because the density ratio does not exist. To circumvent this issue, we propose several new doubly robust estimators based on different kernelization approaches. We analyze the asymptotic mean-squared error of each of these under mild rate conditions for nuisance estimators. Specifically, we demonstrate how to obtain a rate that is independent of the horizon length.




Has companion code repository: https://github.com/CausalML/DoublyRobustOPEForDeterministicPolicies








This page was built for publication: Doubly Robust Off-Policy Value and Gradient Estimation for Deterministic Policies

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6342251)

Retrieved from "https://mardi.schubotz.org/w/index.php?title=Publication:6342251&oldid=37825427"
Powered by MediaWiki