Deconfounding Reinforcement Learning in Observational Settings

Abstract

We propose a general formulation for addressing reinforcement learning (RL) problems in settings with observational data. That is, we consider the problem of learning good policies solely from historical data in which unobserved factors (confounders) affect both observed actions and rewards. Our formulation allows us to extend a representative RL algorithm, the Actor-Critic method, to its deconfounding variant, with the methodology for this extension being easily applied to other RL algorithms. In addition to this, we develop a new benchmark for evaluating deconfounding RL algorithms by modifying the OpenAI Gym environments and the MNIST dataset. Using this benchmark, we demonstrate that the proposed algorithms are superior to traditional RL methods in confounded environments with observational data. To the best of our knowledge, this is the first time that confounders are taken into consideration for addressing full RL problems with observational data. Code is available at https://github.com/CausalRL/DRL.

30 pages

References in corpus (3)

• Continuous State-Space Models for Optimal Sepsis Treatment - a Deep Reinforcement Learning Approach
• Deep Counterfactual Networks with Propensity-Dropout
• Deep Reinforcement Learning for Sepsis Treatment

Cited by in corpus (6)

• Causally Correct Partial Models for Reinforcement Learning
• Disentangled State Space Representations
• Improving Generalizability of Fake News Detection Methods using Propensity Score Matching
• Provably Efficient Causal Reinforcement Learning with Confounded Observational Data
• Off-policy Evaluation in Infinite-Horizon Reinforcement Learning with Latent Confounders
• Environment Reconstruction with Hidden Confounders for Reinforcement Learning based Recommendation