it's probably doable. The main thing you have to do is to implement
ScoreComponentCollection::UnregisterScoreProducer(this);
which is the opposite of
ScoreComponentCollection::RegisterScoreProducer(this);
there are other things you would need to do to get the decoding algorithms
to use the new phrase table. We can thing about those later.
however, i don't understand why you'll want to do this. Phrase-tables can
be binarized so they don't use any RAM, but they just memory map them.
If they use lots of RAM, then it's likely to take a long time to load and
unload so you do want to waste time creating/destroying them too often.
if you have problems implementing it, please let me know & i'll see if i
can help
On 4 December 2013 18:00, Калинин Александр <[email protected]> wrote:
> Many thanx to your replies!
>
> One more question concerning these scripts: are there capability to detach
> translation models to free RAM and attached new TMs while the decoder is
> running? If not maybe you can provide a roadmap for me to contribute such
> functionality?
>
> Kind regards!
>
> 04.12.2013, 18:39, "Rico Sennrich" <[email protected]>:
> > Калинин Александр <verbalab@...> writes:
> >
> >> Hello, everyone!
> >>
> >> I have the following question: I have trained a huge translation model
> >
> > trained on 1m sentences of news
> >
> >> texts. Also I have several minor translation models trained on texts of
> >
> > different domains. Are there any
> >
> >> tools in Moses that can enable merging translation models? Is it
> possible
> >
> > to merge models when the decoder
> >
> >> is running?
> >>
> >> Thanks!
> >>
> >> Kind regards, Alex Kalinin.
> >
> > Hi Alex,
> >
> > there are several scripts that allow you to merge multiple translation
> > models: http://www.statmt.org/moses/?n=Moses.AdvancedFeatures#ntoc52
> >
> > if you're interested in merging models at decoding time, you can do a
> > log-linear combination of models (
> > http://www.statmt.org/moses/?n=Moses.AdvancedFeatures#ntoc18 ), or a
> > linear-interpolation / count-based merge (
> > http://www.statmt.org/moses/?n=Moses.AdvancedFeatures#ntoc55 ).
> >
> > All methods allow you to weight the models in order to prioritize
> in-domain
> > models. Check the documentation/literature if you're interested in more
> details.
> >
> > best wishes,
> > Rico
> >
> > _______________________________________________
> > Moses-support mailing list
> > [email protected]
> > http://mailman.mit.edu/mailman/listinfo/moses-support
>
> --
> С уважением, Александр Калинин
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support
>
--
Hieu Hoang
Research Associate
University of Edinburgh
http://www.hoang.co.uk/hieu
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support
