Off-Policy Evaluation with Out-of-Sample Guarantees

We consider the problem of evaluating the performance of a decision policy using past observational data. The outcome of a policy is measured in terms of a loss (aka. disutility or negative reward)...