Learning to Utilize Shaping Rewards: A New Approach of Reward Shaping