How do I get the predicted function value for a noisy function?

[jperry303]

I have a noisy function f(x,*args), and I would like to know the optimal predicted values of the function given the BO model rather than BO.x_opt or BO.fx_opt. For example, in some cases I have many of the same values of X, and BO.fx_opt is the lowest of these values. I would like to know what the model forecasted value is (some average) for the optimal X value. Also, I would like to know the corresponding optimal X value (because in some cases the true solution may be slightly different from the ones tested by the optimizer).

If my objective is an upward sloping line, and my bounds are [0,1], then the X value of 1 should be the true solution, but if I don't have sufficient iterations, then BO.x_opt might be 0.999 instead of 1. Is there a way to have the predictions from the model result in a value of 1 in this case?

[javiergonzalezh]

Hi @rickycds,

If you want to compute that you can do:

BO.model.model.predict(BO.x_opt)

This will take the model in the object and will used to predict the output in the current optimum. Note, however, that during the optimization, we are already taking into account cases with noisy observations as we are using the get_fmin() function in the models

Hope this helps.