Question on model

[mg10011]

Matt,

I am still fiddling around with your model, and I have a few questions for you if you could be so nice as to clarify.

1. One thing I noticed was that after the inference step, if I go to the AR-HMM model object and try to access the transition matrix that represents the states, I get a full transition matrix that actually has non-zero probabilities of going to states that were not in the model.stateseqs list. Both model.trans_distn.trans_matrix and model.trans_distn.full_trans_matrix are the same, even though the inference procedure "found" a number of states that were far less than Nmax.
2. When using the AR-HSMM model, is there a way to use the PoissonDuration as opposed to the NegativeBinomialDuration? Do you have an opinion as to what sort of prior information or intuition would lead you to use one duration distributional form versus another?

Thanks in advance,
mg

[vaxherra]

Hey, I concur with the point (1). I think those two matrices should be different when Nmax differs from the final number of inferred states.

Could you please investigate?