hmmtrain without updating emission probabilities

I have seen the documentation for the hmmtrain method https://www.mathworks.com/help/stats/hmmtrain.html but I am wondering: is it possible to update the transition probabilities using Baum-Welch while keeping the emission probabilities constant? I would like to keep the emission probabilities the same as they were for the initial guess?

Answer 1

[Paras Gupta]

Hi,

The Baum-Welch algorithm is designed to estimate both the transition and emission probabilities based on the observed sequence. It iteratively updates both sets of probabilities to maximize the likelihood of the observed data.

The MATLAB function hmmtrain can be trained either using the original Baum-Welch Algorithm or the original Viterbi Algorithm. It cannot be customized to keep the emission probabilities constant as that would require a change in the algorithm itself.

If you specifically want to keep the emission probabilities constant and update only the transition probabilities, you will need to implement your own version of the Baum-Welch algorithm.

Hope this helps.