I created the following non-homogeneous Hidden Markov Model using depmix:
#primary time series
datats<-ts(data$primary)
y<- as.numeric(datats)
#Preparing covariates
cov1ts<-ts(data$cov1)
x1<- as.numeric(cov1ts)
cov2ts<-ts(data$cov2)
x2<- as.numeric(cov2ts)
#Build model
hmm_model <- depmix(y~1, data = data.frame(y), nstates = 2, transition = ~ (scale(x1)+scale(x2)))
hmm_model <-fit(hmm_model)
summary(hmm_model)
I now want to make a prediction about the next state. In the past I did this using homogeneous HMM as explained in this post: How to predict out-of-sample observations with depmixS4 package in R?
Specifically, in my case I did:
#[...] Created homogeneous model "hom_model" like before but without transition parameter
#transition probabilities at states
mat1<-hom_model@transition[[1]]@parameters$coefficients
mat1<-hom_model@transition[[2]]@parameters$coefficients
#transition matrix
transmat<-rbind(mat1, mat2)
# prediction as described in post, not very relevant for this question
But now for non-homogeneous hmm, I cannot obtain the transition matrix in the same way because now when I obtain mat1 and mat2, I get the coefficients of the covariates and intercept for each state. Specifically, my output for mat1 in the non-hom case looks like this:
St1 St2
(Intercept) 0 -0.6704946
scale(x1) 0 -1.7279190
scale(x2) 0 -2.0905961
I am unsure on how to obtain the transition matrix for the non-homogeneous case, and also a bit confused as why the State 1 coefficients are all 0.
Thank you
