Adaptive Segmentation of the State Space based on Bayesian Discrimination in Reinforcement Learning