Output from noisy optimization in skopt.gp_minimize

Question

When using skopt.gp_minimize on a noisy dataset with unknown variance - is the returned minimum the x-values found for one specific sample of data or the minimum of the surrogate function? And either way - is it possible to specify which one is desired?

I have looked through the documentation but have been unable to find anything describing this.

Answer 1

It returns the solution with the lowest value of your objective function. To get the expected minimum, use skopt.expected_minimum() (documentation here); more information in this issue.

from skopt import gp_minimize, expected_minimum
res = gp_minimize(...)
print("Best expected:", expected_minimum(res, ...))