Hi, what other kernals commonly used for Gaussian Process are available in Stan besides the exponentiated quadratic kernel? I only see it in the Functions Reference. Thank you.
gp_periodic_cov. I don’t see those in the docs, though they are exposed in the language via the new compiler, @anon79882417 did you ever submit a pr in the docs repo for those?
Please add the doc if not yet done. Things do not exist for users unless documented. I know that this is additional work…but the Adams solver was unknown for years in Stan, for example.
There is an open issue on this in docs: https://github.com/stan-dev/docs/issues/185
Gave it the 2.24 milestone which doesnt magically helps us fix it but it might help us not forget. There is a growing list of stuff that needs to be doc-ed: https://github.com/stan-dev/docs/issues?q=is%3Aopen+is%3Aissue+milestone%3A2.24
Hate to be the great-but-docs-missing-guy but it might be time to start enforcing docs are merged at the same time stanc3 PRs that extend the language are merged. I am definitely guilty of adding stuff that was not doc-ed in time, so I take my share of the blame here.
Thanks for pinging me and for the reminder. It’s on my list. I’ll get to it ASAP. I’m still doing lots of work for school, so apologies for my absence and lack of contributions.
Are these functions in operation at this moment?
May I basically use any of them in the current spot where
If someone can put a quick explanation here for their arguments, it would be fantastic. Thanks.
They are available in the new compiler, but you need to use cmdstan (or cmdstanr/cmdstanpy) as the new compiler is not used by rstan or pystan yet.
If they make sense to use then, yeah I think so? I at least don’t see why not
Yep! Kernel operations are valid, too. Summing, hammard product, but within reason! The implementation is naive so it’s possible you can blow out memory if you do too many operations.
Oh. I have only used rstan within Rstudio thus far. Can I just install the cmdstanr package and then do my work within Rstudio? Sorry for asking a stupid question…
Yep and no worries! Though note if you use cmdstanr, rstudio may tell you a function doesn’t exist etc. when it does the dynamic program checking since rstudio uses rstan to check if functions exist