Marcin, That is exactly the use case that I have in mind. I appreciate the offer. I'll take a look and keep you in the loop.
Cheers, Lane On Fri, Jun 13, 2014 at 10:05 AM, Marcin Junczys-Dowmunt <[email protected]> wrote: > Hi Lane, > We at WIPO implemented a multi-threading and thread-safe version of this > years ago in our branch (wipo). The modifications to the code back then > where quite significant, so we never merged back to master. However if > you need help I can share expierience and coding power. It would help us > to migrate to master withouth loosing this functionality. > > Back then, I bascially implemented a thread-local StaticData god-class. > This allowed to change weights and settings on sentence-base with XML-tags > > <specOpt tm="1 0 1 0 1"/> Translate me > > Unfortunately this code is now not compatible with the current master. > Best, > Marcin > > W dniu 13.06.2014 15:50, Lane Schwartz pisze: >> Thanks, Philipp. That's exactly the info I needed. >> >> On Fri, Jun 13, 2014 at 9:47 AM, Philipp Koehn <[email protected]> wrote: >>> Hi, >>> >>> deep down in ScoreComponentCollection.cpp is the weighting of feature >>> scores: >>> >>> GetWeightedScore() const >>> { >>> return >>> m_scores.inner_product(StaticData::Instance().GetAllWeights().m_scores); >>> } >>> >>> So here you have the weights coming from the global StaticData. >>> >>> This should make instead reference to a weight setting that should be >>> associated >>> with the sentence-specific Manager object. >>> >>> Unfortunately, this function is called at various points in the code where >>> there is >>> no apparent current Manager object, so this would require a bit more >>> reorganization >>> of the code to make it work. >>> >>> If you have time to do that, it would be a useful change. >>> >>> -phi >>> >>> >>> On Fri, Jun 13, 2014 at 9:34 AM, Lane Schwartz <[email protected]> wrote: >>>> Looking at the Alternate Weight Settings section of the documentation, >>>> there is a note "this functionality currently does not work with >>>> multi-threaded decoding." >>>> >>>> >>>> http://www.statmt.org/moses/?n=Moses.AdvancedFeatures#alternateweightsettings >>>> >>>> What is currently preventing this from working in multithreaded mode? >>>> What changes would be required? >>>> >>>> Thanks, >>>> Lane >>>> _______________________________________________ >>>> Moses-support mailing list >>>> [email protected] >>>> http://mailman.mit.edu/mailman/listinfo/moses-support >>> >> >> > > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support -- When a place gets crowded enough to require ID's, social collapse is not far away. It is time to go elsewhere. The best thing about space travel is that it made it possible to go elsewhere. -- R.A. Heinlein, "Time Enough For Love" _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
