Sorry for the naive question, but how can I adjust the Bayesian optimization example shown here to accomodate multidimensional optimization? I understand how to set a Gaussian Process prior for functions which take in multiple inputs, but I’m not sure how to adjust the find_a_candidate
function in the above example.
Hi @hrhm, I guess you want to ask about how to define constraint for multidimensional input? The method in the tutorial can work with the following constraints in PyTorch. If your domain is not available in that list, then you have to write a custom transform which transforms an input x
(in your constrained domain) to a tensor in unconstrained domain and reversely. This is necessary because PyTorch LBFGS does not support constraints.