Add ac function choice, ac function parameters and gaussian process parameters for getting suggestions when using Bayesian Optimization?

[zrcahyx]

This repo https://github.com/fmfn/BayesianOptimization is a good python implementation of BO, but unfortunately it only supports Double type inputs without support for Integer、Discrete and Categorical data types. So i came here for solutions, but unfortunately again, advisor currently can't assign specific acquisition function to use and can't assign parameters for GaussianProcessRegressor from sklearn either. Are you considering adding those features recently?

[tobegit3hub]

Thanks for reporting and it is really appreciated to add these features for Advisor.

In order to simplify the implementation and user's interface, we choose the classic UCB acquisition function. You can change the implementation there or add more acquisition functions as an optional parameter. I will add this feature soon or you can send pull-request if you're interested.

I will look at GaussianProcessRegressor to see if it's possible to be used in this project.

[michellemay]

Could also be interesting to use LBFGSB with multiple restarts to find global maximum of the acquisition function instead of doing thousands of uniform sampling.