Satisficing Q-learning: efficient learning in problems with dichotomous attributes