(I think) I'd like to use the hmm.discnp package for a simple discrete, two-state HMM, but my training data is irregularly shaped (i.e. the observation chains are of varying length). Additionally, I do not see how to label the state of the observations given to the hmm() function. Ultimately, I'd like to 1) train the hmm on labeled data, 2) use viterbi() to calculate optimal labeling of unlabeled observations. More concretely, I have labeled data that looks something like: 11212321221223121221112233222122112 AAAAAAAAABBBBBBBBBBBBBAAAAAAAAAAAAA 21221223121221112233222122112 AAAAAAAAABBBBBBBBBBBBBAAAAAAA 3121221112233222122112 BBBBBBBBBBBBAAAAAAAABB from which I'd like to build the two hidden state (A and B) hmm that emits observed 1, 2, or 3 at probabilities dictated by the hidden state, with transition probabilities between the two states. Given the trained HMM, I then wish to label new sequences via viterbi(). Am I missing the purpose of this package? I also read through the msm package docs, but my data doesn't really have a time coordinate on which the data should be "aligned". Thanks for any pointers, -Aaron <amackey@virginia.edu> [[alternative HTML version deleted]]