Which Explanation Should I Choose? A Function Approximation Perspective to Characterizing Post Hoc Explanations