Useful Policy Invariant Shaping from Arbitrary Advice