function from the LM toolkit abstraction layer. There are two different groups
of classes now:
- LanguageModel, which inherits from StatefulFeatureFunction and contains
the n-gram model feature function.
- LanguageModelImplementation, which is the base class of the individual
LM implementations (SRI, IRST, RandLM, KenLM) and provides methods to
query LM probabilities and states.
Each LanguageModel controls a LanguageModelImplementation. Implementations can
be shared by more than one LanguageModel.
This should make it easier to use the LM libraries as a backend for other
feature functions while retaining the flexibility to use different LM toolkits.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3719 1f5c12ca-751b-0410-a591-d2e778427230
to support multiple translation, language and generation models within the
same process. The main design change is the introduction of a TranslationSystem
object to manage the models, which have been moved out of StaticData.
The changes should have no effect on existing systems.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3394 1f5c12ca-751b-0410-a591-d2e778427230
error catching/fail more gracefully on tables/lm load error & consistent user output
consistent debugging output
cleaned up timing functions
cleaned up mose/moses-cmd api calls/interaction
split up loading of all data in StaticData into separate fns
got binary phrase table to work under WIN32 & passed regression !!
added more comments
deleted phrase table filtering code
deleted mysql support
change calls to ToString() which might affect decoding to a call to a non-debugging fn instead, eg GetString()
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@988 1f5c12ca-751b-0410-a591-d2e778427230