crown421 / gpdiffeq.jl Goto Github PK

This issue is used to trigger TagBot; feel free to unsubscribe.

If you haven't already, you should update your TagBot.yml to include issue comment triggers.
Please see this post on Discourse for instructions and more details.

If you'd like for me to do this for you, comment TagBot fix on this issue.
I'll open a PR within a few hours, please be patient!

Some code that is needed for this package is best placed in other packages. Once the items below are implemented and merged, this package will bring them together:

• Implement/ Fix UniformGrid in InducingPoints.jl
• Identify options for training with Flux (see here
• Implement Generative Models (FITC) in ApproximateGPs.jl

Hello, I love this project and your related Juliacon presentation. I was playing around and notice I am able to construct GPODEFunctions from PosteriorGPs, but the moment that I wrap that posteriorGP in a Stheno atomic() constructor I am no longer able to make a GPODEFunction. This is in spite of the fact that at this point it is identical to the original GP, just with an additional bookeeping GPC() object. Would there be much required to build in Stheno compatibility?

For context, my original intention was to use your pullEuler solver on a Stheno DerivedGP to get a feel for how uncertainty propagates. Alternately, I wonder if perhaps the functionality already exists in some manner not requiring Stheno and I'm missing it. Specifically, is there a way to use this package in the context of universal differential equations where the GP supplies some of the terms but other terms are mechanistic? This could include non-additive interactions, for instance dx = (a + NN(x))*x. I was using Stheno for these, but maybe there is a way to implement this using just AbstractGPs in a way that plays well with your work.