Probability of improvement (PI), Expected Improvement (EI) and Upper Confidence Bound (GP-UCB) acquisition function for Bayesian optimization
Implemented as described in:
E Brochu, VM Cora, & N de Freitas (2010):
“A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and
Hierarchical Reinforcement Learning.”
arXiv:1012.2599, http://arxiv.org/abs/1012.2599