The Parallel Knowledge Gradient Method for Batch Bayesian Optimization