Obtaining a distribution over possible minima of a function

[Ruta Desai]

Hi there,

Thank you for the library and the support. I am trying to use this library for optimization of a high-dimensional continuous function in one of my research projects. I would also like to get an approximate idea of distribution of minima of this function (that I am trying to optimize), after every step of bayesian optimization. Based on this paper: https://www.cs.ox.ac.uk/people/nando.defreitas/publications/BayesOptLoop.pdf (page 12), I thought I could use thompson sampling as acquisition function and since it models the posterior distribution over the unknown minimizer x*, if I sample it n times and average these samples, I will have an empirical estimate (mean) of the possible minima. How can I sample acquisition function (after setting criteria to be thompson sampling?)

Thank you very much for your help!

Ruta

[Ruta Desai]

I meant sample the posterior model. I don't see a sample function for the posterior.

Thanks.

[Ruben Martinez-Cantin]

Hi Ruta,

The Thompson sampling criteria in the library is not a proper Thompson
sampling. In BayesOpt the samples for each x are generated
independently, while in the literature, they are assumed to be
correlated.

You can find many tutorials on how to generate sample functions from a
GP, for example in the tutorial of the GPML. If you make the code,
I'll be happy to integrate a proper Thompson sampling in BayesOpt.

Best,

Ruben

> --
> You received this message because you are subscribed to the Google Groups
> "BayesOpt discussion" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to bayesopt-discus...@googlegroups.com.
> To post to this group, send email to bayesopt-...@googlegroups.com.
> Visit this group at https://groups.google.com/group/bayesopt-discussion.