Doubly robust off-policy evaluation with shrinkage