Thanks for the report -- do I understand correctly that things work fine so long as everything was fit with the latest version, but the latest version of ctKalman does not play nicely with fit objects from earlier ctsem versions? If that's the case then I don't think I can fix this, but in at least some cases like this the ctFitUpdate function is helpful for bringing an older fit object in line with newer version internals.

from ctsem.

Hi Charles,

Looked into it a bit more, ctKalman does not play nicely with fits from the later versions either. Here is a reproducible example using the chicken data from the readme (Note the problem happens when the chicken id does not map directly to the numeric id map). Again, I think this issue is caused by the numeric column selection mentioned above.

`library(data.table)
library(mltools)
library(ctsem)

Runs Fine

chickdata <- one_hot(as.data.table(ChickWeight),cols = 'Diet')

chickdata$weight <- scale(chickdata$weight)

m <- ctModel(
LAMBDA=diag(1), #Factor loading matrix of latent processes on measurements, fixed to 1
type = 'stanct', #Could specify 'standt' here for discrete time.
tipredDefault = FALSE, #limit covariate effects on parameters to those explicitly specified
manifestNames='weight', #Observed measurements of the latent processes
latentNames='Lweight', #Names here simply make parameters and plots more interpretable
TIpredNames = paste0('Diet_',2:4), #Covariates, in this case one category needs to be baseline...
DRIFT='a11 | param', #normally self feedback (diagonal drift terms) are restricted to negative
MANIFESTMEANS=0, #For identification CINT is normally zero with this freely estimated
CINT='cint ||||Diet_2,Diet_3,Diet_4', #diet covariates specified in 5th 'slot' (four '|' separators)
time='Time',
id='Chick')

f <- ctStanFit(chickdata,m,nopriors=FALSE)

ctKalman(f,plot=TRUE,subjects=2:4,kalmanvec=c('yprior','ysmooth'))

Breaks if the chicks have character ids

chickdata <- one_hot(as.data.table(ChickWeight),cols = 'Diet')
chickdata$weight <- scale(chickdata$weight)
chickdata$Chick <- paste0(chickdata$Chick, "a") # Create Chick character id

m <- ctModel(
LAMBDA=diag(1), #Factor loading matrix of latent processes on measurements, fixed to 1
type = 'stanct', #Could specify 'standt' here for discrete time.
tipredDefault = FALSE, #limit covariate effects on parameters to those explicitly specified
manifestNames='weight', #Observed measurements of the latent processes
latentNames='Lweight', #Names here simply make parameters and plots more interpretable
TIpredNames = paste0('Diet_',2:4), #Covariates, in this case one category needs to be baseline...
DRIFT='a11 | param', #normally self feedback (diagonal drift terms) are restricted to negative
MANIFESTMEANS=0, #For identification CINT is normally zero with this freely estimated
CINT='cint ||||Diet_2,Diet_3,Diet_4', #diet covariates specified in 5th 'slot' (four '|' separators)
time='Time',
id='Chick')

f <- ctStanFit(chickdata,m,nopriors=FALSE)

ctKalman(f,plot=TRUE,subjects=2:4,kalmanvec=c('yprior','ysmooth')) `

from ctsem.

I see, thanks a lot. It's not 'strictly' an error but rather just poor coding practice re backwards compatibility -- in the 2nd example, you really need to specify subjects=c('2a',3a','4a'), but that doesn't match up with earlier versions so I've pushed a fix that allows both specifications (and warns if integers are used when realid=TRUE, which is the default)

from ctsem.

This happens when I use model objects fit with older versions of ctsem. Maybe the id mapping changed? Instead of selecting by column number maybe it should be by column name.

from ctsem.