Approximate Policy Iteration with Demonstration Data