Timezone: »

Off-Policy Evaluation via the Regularized Lagrangian
Mengjiao (Sherry) Yang · Ofir Nachum · Bo Dai · Lihong Li · Dale Schuurmans

Mon Dec 07 09:00 PM -- 11:00 PM (PST) @ Poster Session 0 #161

The recently proposed distribution correction estimation (DICE) family of estimators has advanced the state of the art in off-policy evaluation from behavior-agnostic data. While these estimators all perform some form of stationary distribution correction, they arise from different derivations and objective functions. In this paper, we unify these estimators as regularized Lagrangians of the same linear program. The unification allows us to expand the space of DICE estimators to new alternatives that demonstrate improved performance. More importantly, by analyzing the expanded space of estimators both mathematically and empirically we find that dual solutions offer greater flexibility in navigating the tradeoff between optimization stability and estimation bias, and generally provide superior estimates in practice.

Author Information

Mengjiao (Sherry) Yang (Google Brain)
Ofir Nachum (Google Brain)
Bo Dai (Google Brain)
Lihong Li (Amazon)
Dale Schuurmans (Google Brain & University of Alberta)

More from the Same Authors