[Biojava-l] Partial path probs

[Matthew Pocock]

Schreiber, Mark wrote:

> Hi -
> 
> Is there a way to get the partial path probabilities for the viterbi
> algorithm in DP (eg the path trellis)?

Well, given a state path, the score for the region i..j is sum(score(x)) 
for all x in [i..j]. You can call the dp code in such a way that it 
writes the DP matrix to an array rather than using the space-saving 
implementation (DPCursor).

Also, we could change the viterbi() method interface so that it takes a 
call-back that will be informed of each path. This could be informed of 
every path (including sub-optimal ones) as the DP matrix is built. That 
would mean somebody doing DP code-mining.

> 
> I suppose I could subclass it, perhaps VerboseDP :)
> 
> On a more theoretical note if one calculated the probability of the
> viterbi path and compared that to the forward probability would that be
> a good way of infering confidence in the predictions fit to the model?
> Eg If the Viterbi path prob and the Forward prob were close you could be
> confident that, according to your model, other possible paths are not
> that likely. Alternatively if they are not close you might conclude that
> although the viterbi path is the most parsimonious there could be other
> paths that are almost as likely. Or am I barking up the wrong tree here?

This sounds sane. If the backwards score is very much higher than the 
viterbi score then there are many paths that score 'well enough'. The 
best thing to do is to display the posterior decoding 
(forwards-backwards matrix) and overlay your viterbi path on that. This 
will tell you exactly which bits of the alignment are dodgey.

Matthew

> 
> - Mark
> =======================================================================
> 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.
> =======================================================================
> 
> _______________________________________________
> 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