Off-policy evaluation uses logged actions, outcomes, and selection probabilities to estimate a policy not yet deployed. Poor coverage makes estimates unstable or impossible.

Off-Policy Evaluation Estimates a New Policy From Old Logs

Algorithms can infer candidate causal structures from observational or interventional data, but equivalence classes, hidden confounders, and sampling choices limit conclusions.

Causal Discovery Produces Hypotheses Under Assumptions

In this situation—estimating a new tutor router before exposing learners—which choice best applies “Off-Policy Evaluation Estimates a New Policy From Old Logs”?

Learn Before

Related