Possible ways to speed up spatio-temporal GP modeling

Questions v5
14

Wow, thank you so much! I am getting a lot of help today. I really appreciate your time!

I love your statement:

It will give you the same posterior of f as if you used the GP to “predict” what the f values would be there.

It is really great that we do get predictions while doing inference!

For the time, we are doing hourly modeling by month for the entire 2020. So, for a month, the full time index will be a size of 24*30.

For the exact spatial domain boundary, I’d have to ask my postdoc, but it is roughly the Southern California area (LA County + suburban areas). The satellite resolution is ~ 2 km. My postdoc is using an analytical solution, and, in parallel, I am now into PyMC GP!

Thank you!