How does HSGP fit into this library (if it does)

[ricardoV94]

code search didn't yield anything

[bwengals]

I'm not sure, a couple options. I pictured this as depending on pymc pretty much just for setting priors. But having all the kernels here and in PyMC is super redundant. Options:

1. Put all GP stuff here, including MCMC targeted approaches like HSGP.

2. Keep just HSGP in PyMC. Remove all the other GP stuff. You don't need all the covariance matrix apparatus for HSGPs, just expressions for the power spectral density. If someone wants to use a vanilla GP or one of the basic inducing point ones within a PyMC model and use NUTS they could import those bits from ptgp.

3. Keep HSGPs and other MCMC appropriate models in PyMC. Vanilla GPs and the basic inducing point ones work.

[jessegrabowski]

fwiw i thought your argument that HSGP goes in pymc because it's MCMC/prior focused was convincing. This repo is for "I just want a GP model", which is not really the hsgp use-case.

[ricardoV94]

Is HSGP only useful for mcmc?