[Biojava-dev] ModelInState fixed?

[Matthew Pocock]

Mm.

Anybody would think that somebody other than Thomas
and me acutaly uses the dp package!

Ok - firstly, the ModelInState stuff is a realy nasty
hack, and appears currently to not work very well.
I've not used it in quite a while, so haven't been
supporting it. But, if you want it working, we can
look together at fixing it (hint hint).

Now for the next one - states that emit more than one
symbol. At the moment, one state emits 1 symbol at a
time. This makes the code simple. It sucks for things
like aligning dna to protein as the DNA inserts want
to be nucleotides but the dna-protein matches want to
be codons. This can be fixed. The advance arrays don't
need to contain values of just 0 or 1 - they could for
example be 3. This has a knock-on for the emission
alphabet in that now it emits both nucleotides and
trinucleotides, but that's fixable. To make this work,
we need to update the DP cursors to be aware that they
have to store more than just the last one column.

Now for the last one - states with duration. I've
never done this. As far as I know, this is handled by
having a duration dimension to the DP matrix as well
as the sequence dimension(s) and state dimension. If
anybody has a paper/book that tells us how to do this,
I'd be happy to work with someone to implement it.

Matthew

 --- Russell Smithies <russell.smithies at xtra.co.nz>
wrote:
> 
> 
> Hi,
> 
> Did the HMM ModelInState stuff ever get fixed?
> (the bit Matthew broke:-)
> 
> Russell
> 
> > Message: 2
> > Date: Mon, 14 Apr 2003 10:57:05 +1200
> > From: "Schreiber, Mark"
> <mark.schreiber at agresearch.co.nz>
> > Subject: RE: [Biojava-dev] bytecode
> > To: "Matthew Pocock" <matthew_pocock at yahoo.co.uk>,
> > <biojava-dev at biojava.org>
> > Message-ID:
> >
>
<AF026AF0FF4B054590228FD1F1DE5165011BA438 at inbox.agresearch.co.nz>
> > Content-Type: text/plain; charset="utf-8"
> >
> > Hi -
> >
> > Can I put in a request for semi hidden markov
> models or states with a
> "time spent in state" distribution if your not too
> busy :)
> >
> > - Mark
> >
> >
> > -----Original Message-----
> > From: Matthew Pocock
> [mailto:matthew_pocock at yahoo.co.uk]
> > Sent: Mon 14/04/2003 9:25 a.m.
> > To: biojava-dev at biojava.org
> > Cc:
> > Subject: [Biojava-dev] bytecode
> >
> >
> >
> > Hi,
> >
> > I've been poking arround in the bytecode package.
> > CodeGenerator now has two methods - stackDepth()
> and
> > stackDelta() that let us calcualte a stack depth
> for
> > methods (or macros). This lets us chose a sane
> stack
> > depth for methods when making the class files -
> > hopefully removing David's errors.
> >
> > BioJava compiles against this new bytecode.jar,
> but
> > could  people take this for a spin before we
> commit it
> > to biojava-live - I don't want to break all our
> > feature projection stuff by accident.
> >
> > If I get bored this comming week, I'll add in full
> > support for generics / templates to the package.
> This
> > should let us define algorithms like "sort" for
> any
> > type and then emit the bytecode for the algorithm
> over
> > a concrete type like "int". Should come in usefull
> in
> > like a million little places - like in DP kernels,
> or
> > SSAHA inner-loops, or turning math data-structures
> > into code.
> >
> > Matthew
> >
> 
> 
> _______________________________________________
> biojava-dev mailing list
> biojava-dev at biojava.org
> http://biojava.org/mailman/listinfo/biojava-dev 

__________________________________________________
Yahoo! Plus
For a better Internet experience
http://www.yahoo.co.uk/btoffer