Decreasing likelihood

[edvinli]

When running hmmDemo() and changing model.observationDimension to 1 (instead of 10) and observation length to T = 100 I get that the likelihood is decreasing, when it always should be increasing right? The result can be seen in the attached figure. Does mean that the EM algorithm in hmmFit has an error, or am I missing something?

[edvinli]

Hm, correct me if I am wrong but perhaps what is printed in the figure above is the loglikelihood without the log(priors)? So what the EM guarantees is that loglik + log(prior) should increase, but loglik could actually decrease?