Tactical Optimism and Pessimism for Deep Reinforcement Learning