Benchmarking Deep Multi-Agent Reinforcement Learning Algorithms