Possible ways to speed up spatio-temporal GP modeling

Thanks for the suggestion! By the way, I saw the new state space (SS) model implementation. I’ve used state space models in R for a while, primarily simple random walk models. I was delighted to see your developing SS models within the PyMC framework.

I guess my raster cells are too many, although someone in my group is testing spatial aggregation before doing any GP or analytical multivariate normal modeling. But your idea of a diagonal VAR sounds very interesting.

I will keep paying attention to your SS models.

Thank you very much!

2 Likes