I believe i’ve fixed my own issues.
I used a softmax to force the contraint that rows of p_trans sum to one.
I also realized i need better priors on alpha, so i don’t get wild aphas in the first place. I’ll post the complete solution here when I fix the kinks.
1 Like