Adaptive Traffic Control with Deep Reinforcement Learning: Towards State-of-the-art and Beyond