is there a more detailed explanation on the use of bayes in hyperparameter search? For example, if it uses GP regression, what are the default kernel and parameters, and how to change them, etc.
Hi @zyzhang , could you expand on your question regarding Bayes. Are asking to have more hands on modification of our implementation? Unfortunately we don’t support deeper control options at this time, but I am curious about your use case here.
Please visit this detailed article on the specifics of how Bayesian optimization works.
@mohammadbakir I see. I was hoping to see at least something like, if GP is used, even if I can’t change the kernel, I should be able to tell what kernel, scale length (and other parameters) are used, so the whole thing is not just a blackbox to me.
Thank you for the feedback @zyzhang. We are working on expanding our sweeps features functionality. I will keep note of your comments and add them to our DB. Once there has been movement on this feature, I will update you.
Sorry one more clarification. the ‘bayes’ search method apparently supports discrete search as well. May I know how is it done since by default GP shouldn’t be able to do discrete search?