Thompson Sampling for Parameterized Markov Decision Processes with Uninformative Actions