The knowledge gradient for sequential decision making with stochastic binary feedbacks