Hi,
Thanks for linking the repo! I see from @betanalpha’s reply to @mbjoseph’s post (Are time-varying transition probabilities supported by the new hmm_* functions? - #3 by martinmodrak) that time-varying transition matrices are coming in the feature’s next iteration - is it still worth creating an issue on the repo in this case?
From my understanding (which may be wrong, as I’m pretty new to HMM’s!) we’re wanting to pass an array which looks something like
matrix[K,K] theta[T]
where K is the number of hidden states and T is the number of time steps, with the values of theta at each timestep depending on one or more covariates.
There’s also the issue of having multiple time series (in my case camera trap sites) which are generally not the same length (camera traps may end up being deployed for different numbers of days, especially when in remote areas which take time to access) - I’m guessing that this means we actually need an array which contains a separate transition matrix for each site/time combination (as this is what I did in R when simulating data to test the model on), but I’m not certain.