Hidden Markov Models

[struktured]

I didn't see any HMM implementations yet, any objection if I implement one and add it? It would probably be similar in api to https://github.com/hmmlearn/hmmlearn.

[rleonid]

That would be a welcome addition, go for it.

[superbobry]

I guess I could have a go at this.

Do you prefer MLE, MAP or Bayesian (mean-field) HMMs? hmmlearn implements MAP, but since we're talking about a from scratch implementation here, a proper Bayesian model might be more appropriate.

[rleonid]

I'd welcome any implementation that you find useful and want to share.

[vd1]

Is anyone doing this? I would be interested in using it; Bayesian version the better option for me.

[rleonid]

I am currently not working on this.

[rleonid]

I should also mention that while I'd love for this to a feature of Oml, perhaps leveraging some of the capabilities of Libra might make an implementation easier if you're looking for a pure OCaml, Bayesian, solution.