One of the common ways for showing the trade_off between exploration_exploitation in reinforcement learning problems is the multi_armed bandit problem. In this paper we consider the MABP in a nonstationary environment which features change during the period of learning. The represented learning algorithms are intuition based solutions to the exploration_explotation tarde_off that are called ad hoc method. These methods include action_value methods with e-greedy and softmax action selection rules, the probability matching method and finally the adaptive pursuit method. For producing near optimal results we change the ad hoc methods to sequential optimistic ad hoc methods which provide us completely better results.

Keywords