Quantifying Differences in Reward Functions