search

NIEHS / PrestoGP

Penalized Regression on Spatiotemporal Outcomes using Gaussian Processes a.k.a. PrestoGP
https://niehs.github.io/PrestoGP/
0 stars 0 forks source link

Full likelihood is not matching the Vecchia likelihood in some cases #18

Closed ericbair-sciome closed 7 months ago

ericbair-sciome commented 7 months ago

As I understand it, the maximum Vecchia likelihood should equal the full likelihood when m=n-1, where m is the number of nearest neighbors and is the number of observations. This does not seem to be the case for some examples I have tested. I suspect the full multivariate likelihood function may have issues, given that we have not tested (or even used) this function in the past. I will confirm and fix.

kyle-messier commented 7 months ago

@ericbair-sciome Thanks for adding all of the tasks from the Teams kanban board here. As the issues are labeled, some of these can wait for version 2.0 or are otherwise "not-needed, but nice to have" kind of things. I'll go through an add/update labels. Please feel free to edit said labels if you plan to do it for v1.0, but it is labeled otherwise. @ericbair-sciome @brian-bk22 @sciome-bot

ericbair-sciome commented 7 months ago

This should be fixed in the latest update on the main-sciome branch. I created a pull request from that branch to the main branch. Am I the one who should be creating these pull requests? I guess I will keep doing it when I update things unless someone tells me not to. :)