It seems that the time complexity is O(2^S), where S is num_to_sample

[hzhangxyz]

so it is hard to do some parallel computing, is there some solution?

besides, could you mention it in the document? not only "O(N^3), where N is the number of historical points"

[wujian16]

I have not understood where O(2^S) comes from, please refer to the paper "Parallel Bayesian Global Optimization of Expensive Functions" for details.

At the same time, you should try running the code to appreciate how fast the code can be for the number of evaluations up to 500 ( and num_to_sample up to 20).