Associative methods in reinforcement learning: an empirical study