Simplifying Deep Temporal Difference Reinforcement