From: Philipp K. <pk...@in...> - 2007-07-19 19:13:06
|
Hi, further comment: the original binary ttable code and binary reordering model code cached at several places the full unpruned phrase table entries for a source phrase. For a Europarl model this means storing over 100,000 translations for the comma ",". Now, only the filtered entries are stored (no repeated lookups are necessary for multiple commas). -phi On 7/19/07, ph...@us... <ph...@us...> wrote: > Revision: 1428 > http://mosesdecoder.svn.sourceforge.net/mosesdecoder/?rev=1428&view=rev > Author: phkoehn > Date: 2007-07-19 12:03:40 -0700 (Thu, 19 Jul 2007) > > Log Message: > ----------- > fixed memort blowup with caching bin ttable / bin reordering table > > Modified Paths: > -------------- > trunk/moses/src/Hypothesis.cpp > trunk/moses/src/LexicalReorderingTable.cpp > trunk/moses/src/PDTAimp.h > trunk/moses/src/PhraseDictionaryTreeAdaptor.cpp > trunk/moses/src/TranslationOptionCollection.cpp > > > This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site. > > ------------------------------------------------------------------------- > This SF.net email is sponsored by: Microsoft > Defy all challenges. Microsoft(R) Visual Studio 2005. > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ > _______________________________________________ > Mosesdecoder-commits mailing list > Mos...@li... > https://lists.sourceforge.net/lists/listinfo/mosesdecoder-commits > > |