Deep Reinforcement Learning for Asset Allocation: Reward Clipping