Both here and in the model with a time-varying mean, the first p values (p is AR order) of the Phi parameter have a really strong prior. I think I originally did this b/c I stopped wanting to use the uniform prior (perhaps b/c of infinite slice? more later), and when I started using the normal, I didn't want it to allow for values outside the -1,1 interval. This must have been before I discovered the truncation notation, as I use this T() notation in other parts of the model.
Whatever the reason, the main motivation behind my close and revisionist inspection of these models is that I am sometimes getting this weird 'infinite' slice error from JAGS --- it's indicating the Phi state is suddenly gaining probability of 1 or something like that. Basically, it is getting stuck.
spolischook
commented
6 years ago
https://github.com/rBatt/timeScales/blob/6ea3a8770c8bd2e47f34872a6e0bc064d9cc04bb/inst/jags/tv_arp_noMean.jags#L45
Both here and in the model with a time-varying mean, the first
pvalues (p is AR order) of the Phi parameter have a really strong prior. I think I originally did this b/c I stopped wanting to use the uniform prior (perhaps b/c of infinite slice? more later), and when I started using the normal, I didn't want it to allow for values outside the -1,1 interval. This must have been before I discovered the truncation notation, as I use this T() notation in other parts of the model.Whatever the reason, the main motivation behind my close and revisionist inspection of these models is that I am sometimes getting this weird 'infinite' slice error from JAGS --- it's indicating the Phi state is suddenly gaining probability of 1 or something like that. Basically, it is getting stuck.
Get back to this later.