Learning When-to-Treat Policies
arXiv:1905.09751
Abstract
Many applied decision-making problems have a dynamic component: The policymaker needs not only to choose whom to treat, but also when to start which treatment. For example, a medical doctor may choose between postponing treatment (watchful waiting) and prescribing one of several available treatments during the many visits from a patient. We develop an "advantage doubly robust" estimator for learning such dynamic treatment rules using observational data under the assumption of sequential ignorability. We prove welfare regret bounds that generalize results for doubly robust learning in the single-step setting, and show promising empirical performance in several different contexts. Our approach is practical for policy optimization, and does not need any structural (e.g., Markovian) assumptions.
References in corpus (12)
- Recursive Partitioning for Heterogeneous Causal Effects
- Meta-learners for Estimating Heterogeneous Treatment Effects using Machine Learning
- Doubly Robust Policy Evaluation and Learning
- Breaking the Curse of Horizon: Infinite-Horizon Off-Policy Estimation
- Reliable Decision Support using Counterfactual Models
- Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes
- Balanced Policy Evaluation and Learning
- Information-Theoretic Considerations in Batch Reinforcement Learning
- Double/Debiased Machine Learning for Treatment and Causal Parameters
- Using Options and Covariance Testing for Long Horizon Off-Policy Policy Evaluation
- Representation Balancing MDPs for Off-Policy Policy Evaluation
- Faster Rates for Policy Learning
Cited by in corpus (5)
- Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems
- Empirical Study of Off-Policy Policy Evaluation for Reinforcement Learning
- Off-policy Policy Evaluation For Sequential Decisions Under Unobserved Confounding
- Statistically Efficient Off-Policy Policy Gradients
- Active Offline Policy Selection