Timezone: »

 
Spotlight
Model Selection in Batch Policy Optimization
Jonathan Lee · George Tucker · Ofir Nachum · Bo Dai

Thu Jul 21 11:10 AM -- 11:15 AM (PDT) @ Room 301 - 303
in Reinforcement Learning »

We study the problem of model selection in batch policy optimization: given a fixed, partial-feedback dataset and M model classes, learn a policy with performance that is competitive with the policy derived from the best model class. We formalize the problem in the contextual bandit setting with linear model classes by identifying three sources of error that any model selection algorithm should optimally trade-off in order to be competitive: (1) approximation error, (2) statistical complexity, and (3) coverage. The first two sources are common in model selection for supervised learning, where optimally trading off these two is well-studied. In contrast, the third source is unique to batch policy optimization and is due to dataset shift inherent to the setting. We first show that no batch policy optimization algorithm can achieve a guarantee addressing all three simultaneously, revealing a stark contrast between difficulties in batch policy optimization and the positive results available in supervised learning. Despite this negative result, we show that relaxing any one of the three error sources enables the design of algorithms achieving near-oracle inequalities for the remaining two. We conclude with experiments demonstrating the efficacy of these algorithms.

Keywords: RL: Batch/Offline · T: Online Learning and Bandits · T: Reinforcement Learning and Planning

Author Information

Jonathan Lee (Stanford University)
George Tucker (Google Brain)
Ofir Nachum (Google Brain)
Bo Dai (Google Brain)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors