Optimization for Reinforcement Learning: From a single agent to cooperative agents