I'm using bayesian-optimization to optimise a black-box function. My function is quite noisy, so I would like to get expected maximum rather than point maximum.
That package uses scikit-learn's Gaussian Process Regressor underneath. In one of the issues they suggest using the expectation surface.
Is there a way to get the surface and the expected maximum directly from scipy?
Currently, I'm just running a mesh grid through the predictor and getting maximum that way, but I was wondering if I can perhaps get it directly.