Estimating probability of next state (time t+1)

[asutosh]

Given a model and observation sequence till time t, how do i compute the probability of next state(s) time t+1 (, t+2 and so on...)?

Does Jahmm library provide this function. If not, can anyone show me how to implement this.

 

 

regards
asutosh

[Sadegh Dorri N.]

Asutosh,

You should first learn an HMM by using the BaumWelch algorithm
(be.ac.ulg.montefiore.run.jahmm.learn.BaumWelchLearner.learn()
or
be.ac.ulg.montefiore.run.jahmm.learn.BaumWelchScaledLearner.learn())

with your observation sequence. Then use the learned HMM to calculate
the probability of the sequence in the learned HMM (Hmm.probability())
and divide it by the probability of the sequence plus the observation
you
want to calculate its probability. If you use scaling, you should use
Hmm.lnProbability() and subtract the probabilities then exponentiate
the result.

Regards,
Sadegh

> --
> You received this message because you are subscribed to the Google
> Groups "Jahmm HMM library" group.
> To view this discussion on the web visit
> https://groups.google.com/d/msg/jahmm-discuss/-/gq70ggCJ3acJ.
> To post to this group, send email to jahmm-...@googlegroups.com.
> To unsubscribe from this group, send email to
> jahmm-discus...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/jahmm-discuss?hl=en.

[Sadegh Dorri N.]

Of course, if you have the model, the learning step is not required.

On Wed 12 Sep 2012 03:42:19 PM IRDT, asutosh wrote: