Reinforcement learning for the traveling salesman problem: Performance comparison of three algorithms