[Biojava-l] Relations between trainers

[Schreiber, Mark]

Hi -

Just to let you know that I have done this for my PhD project (using biojava) and it works remarkably well if your initial parameters are close'ish to the actual 'true model'. It didn't work so well (for what I was doing) if you completely randomised the initial values. A word of warning: Viterbi training is sometimes frowned upon by HMM purists (not sure why).

- Mark


> -----Original Message-----
> From: Matthew Pocock [mailto:matthew_pocock at yahoo.co.uk] 
> Sent: Thursday, 6 March 2003 3:48 a.m.
> To: Oscar Lopes
> Cc: biojava-l at biojava.org
> Subject: Re: [Biojava-l] Relations between trainers
> 
> 
> Hi Oscar,
> 
> For viterbi training, you need to do the following:
> 
> * Create your model & initialize with random parameters
> * Create a hmm training context and register the model with it
> * do
>    o foreach training sequence
>      - get the viterbi aignment of the model to your sequence
>      - add one count to the training context for each time a state or 
> transition is used
>    o call the context's train method
> * while the alignments change
> 
> Good luck. I'd be interested to see what you end up with. If 
> you need a 
> hand with any of the APIs, send questions to the list.
> 
> Matthew
> 
> Oscar Lopes wrote:
> > Hi!
> > 
> > I'm a graduate student and I'm using biojava to implement Viterbi
> > training algorithm.
> > I spent some time digging in the code, but I still can't make the 
> > relation between
> > TrainigAlgorithm, DistributionTrainer, etc...
> > 
> > What are the key classes that I can use? How does it work? 
> I tried to
> > look at
> > BaumWelchTrainer, but it didn't make sense how I would 
> implement Viterbi 
> > Trainig.
> > Besides, I kept getting confused with DistributionTrainer and 
> > DistributionTrainerContext...
> > 
> > Can anybody help me?? It would be nice if someone explain to me how 
> > the
> > whole thing works.
> > 
> > Thanks in advance
> > 
> >                           Óscar Lopes
> > 
> > Ps. Sorry about my english.... :)
> > 
> > oscar at gauss.inesc.pt
> > ombl at rnl.ist.utl.pt
> > 
> > _______________________________________________
> > Biojava-l mailing list  -  Biojava-l at biojava.org 
> > http://biojava.org/mailman/listinfo/biojava-l
> > 
> 
> 
> -- 
> BioJava Consulting LTD - Support and training for BioJava 
> http://www.biojava.co.uk
> 
> 
> _______________________________________________
> Biojava-l mailing list  -  Biojava-l at biojava.org 
> http://biojava.org/mailman/listinfo/biojava-l
> 
=======================================================================
Attention: The information contained in this message and/or attachments
from AgResearch Limited is intended only for the persons or entities
to which it is addressed and may contain confidential and/or privileged
material. Any review, retransmission, dissemination or other use of, or
taking of any action in reliance upon, this information by persons or
entities other than the intended recipients is prohibited by AgResearch
Limited. If you have received this message in error, please notify the
sender immediately.
=======================================================================