Debunking the mysteries of deep reinforcement learning