Extracting combinations of emitted states about seqhmm HOT 4 CLOSED

This is available via mc_to_sc function, which combines a multichannel model to a single-channel model.

from seqhmm.

Hi Jouni and thanks for the response.

Could you please confirm if the multi-channel and single channel models are equivalent or does using the single channel model require a re-estimation of the model parameters? And are the proportions of the slices of the combined states in a directed graph equal to the emission probabilities of the combined states in a single channel?

Thanks for your help!

from seqhmm.

The models are equivalent under the assumption that the channels are independent given the hidden states, in which case we can just multiply the emission probabilities of different channels. But if this assumption does not hold, re-estimating the single channel model can lead to different results (although in practice with complex models it could happen even under the independence assumption due to the multiple local optima in the likelihood surface).

Yes, the slices of combined states in the plot are just sums of the corresponding emission probabilities.

from seqhmm.

Ok, all understood. thank you again for your help.

from seqhmm.