The thing is, I don't know what GPyOpt's 'default' approach is for mixed domains. Are discrete variables being converted into one-hot vectors (a la David Duvenaud's kernel cookbook strategy)? Are discrete variables being treated as continuous variables and then converted to integers internally? Is a Matern52 kernel still being used with mixed domains or is another kernel being used?
Thanks so much for your help. I greatly appreciate it.
Hello.
First off, been loving GPyOpt! So, keep up the great work!
Am trying out a custom kernel for mixed domain optimization and am comparing it to GPyOpt's default approach for mixed domains (referenced in this notebook: https://nbviewer.jupyter.org/github/SheffieldML/GPyOpt/blob/devel/manual/GPyOpt_mixed_domain.ipynb).
The thing is, I don't know what GPyOpt's 'default' approach is for mixed domains. Are discrete variables being converted into one-hot vectors (a la David Duvenaud's kernel cookbook strategy)? Are discrete variables being treated as continuous variables and then converted to integers internally? Is a Matern52 kernel still being used with mixed domains or is another kernel being used?
Thanks so much for your help. I greatly appreciate it.
Michael