CODEWITHSUNDEEP
pythonmachine-learningoptimization
user442920
user442920

Reputation: 847

Max dimension for Bayesian Optimization (GPyOpt, GPFlow)

Does anybody know how quickly the Bayesian Optimization algorithm slows down as a function of the dimension of the search space? What is a good estimate of the maximum dimension that one can reasonably use? I am thinking especially of GPyOpt and GPFlowOpt.

Upvotes: 0

Views: 287

Answers (1)

Learning is a mess
Learning is a mess

Reputation: 8277

As a general rule of thumb, Bayesian optimisation becomes ineffective when the dimension of the search space is > 15. This will obviously depend on volume and utility function landscape. Check https://arxiv.org/abs/1902.10675 for an example of bayesian optimisation coupled with some reversible dimensional reduction (an auto-encoder there).

Upvotes: 1

Related Questions