diff options
51 files changed, 1143 insertions, 384 deletions
diff --git a/compile.sh b/compile.sh index 043f47c30..01f86bc12 100755 --- a/compile.sh +++ b/compile.sh @@ -3,5 +3,6 @@ # you can install all 3rd-party dependencies by running make -f contrib/Makefiles/install-dependencies.gmake set -e -o pipefail -./bjam --with-irstlm=./opt --with-boost=./opt --with-cmph=./opt --with-xmlrpc-c=./opt --with-mm --with-probing-pt -j$(getconf _NPROCESSORS_ONLN) $@ +opt=$(pwd)/opt +./bjam --with-irstlm=$opt --with-boost=$opt --with-cmph=$opt --with-xmlrpc-c=$opt --with-mm --with-probing-pt -j$(getconf _NPROCESSORS_ONLN) $@ diff --git a/contrib/other-builds/moses/.project b/contrib/other-builds/moses/.project index 009acf0ba..6c9b33129 100644 --- a/contrib/other-builds/moses/.project +++ b/contrib/other-builds/moses/.project @@ -2431,6 +2431,16 @@ <locationURI>PARENT-3-PROJECT_LOC/moses/parameters/NBestOptions.h</locationURI> </link> <link> + <name>parameters/OOVHandlingOptions.cpp</name> + <type>1</type> + <locationURI>PARENT-3-PROJECT_LOC/moses/parameters/OOVHandlingOptions.cpp</locationURI> + </link> + <link> + <name>parameters/OOVHandlingOptions.h</name> + <type>1</type> + <locationURI>PARENT-3-PROJECT_LOC/moses/parameters/OOVHandlingOptions.h</locationURI> + </link> + <link> <name>parameters/OptionsBaseClass.cpp</name> <type>1</type> <locationURI>PARENT-3-PROJECT_LOC/moses/parameters/OptionsBaseClass.cpp</locationURI> diff --git a/defer/ExternalFeature.h b/defer/ExternalFeature.h index 6c0fb829e..3755bf4ff 100644 --- a/defer/ExternalFeature.h +++ b/defer/ExternalFeature.h @@ -51,27 +51,8 @@ public: void SetParameter(const std::string& key, const std::string& value); - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedFutureScore) const { - } - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedFutureScore = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - - FFState* EvaluateWhenApplied( - const Hypothesis& cur_hypo, - const FFState* prev_state, - ScoreComponentCollection* accumulator) const; + FFState* EvaluateWhenApplied(const Hypothesis& cur_hypo, const FFState* prev_state, + ScoreComponentCollection* accumulator) const; FFState* EvaluateWhenApplied( const ChartHypothesis& /* cur_hypo */, diff --git a/moses/FF/BleuScoreFeature.h b/moses/FF/BleuScoreFeature.h index 266d06b68..753d92cf9 100644 --- a/moses/FF/BleuScoreFeature.h +++ b/moses/FF/BleuScoreFeature.h @@ -123,22 +123,6 @@ public: FFState* EvaluateWhenApplied(const ChartHypothesis& cur_hypo, int featureID, ScoreComponentCollection* accumulator) const; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const { - } bool Enabled() const { return m_enabled; diff --git a/moses/FF/ConstrainedDecoding.h b/moses/FF/ConstrainedDecoding.h index 3f28c43e1..769edd80f 100644 --- a/moses/FF/ConstrainedDecoding.h +++ b/moses/FF/ConstrainedDecoding.h @@ -42,24 +42,6 @@ public: return true; } - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const { - } - - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/FF/ControlRecombination.h b/moses/FF/ControlRecombination.h index 04f7d441b..034b1a790 100644 --- a/moses/FF/ControlRecombination.h +++ b/moses/FF/ControlRecombination.h @@ -58,23 +58,6 @@ public: return true; } - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const { - } - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/FF/CoveredReferenceFeature.cpp b/moses/FF/CoveredReferenceFeature.cpp index d4606a8ec..5e4ada5b0 100644 --- a/moses/FF/CoveredReferenceFeature.cpp +++ b/moses/FF/CoveredReferenceFeature.cpp @@ -30,11 +30,6 @@ bool CoveredReferenceState::operator==(const FFState& other) const } ////////////////////////////////////////////////////////////////////////////////////////////////////////////// -void CoveredReferenceFeature::EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const -{} void CoveredReferenceFeature::EvaluateWithSourceContext(const InputType &input , const InputPath &inputPath diff --git a/moses/FF/CoveredReferenceFeature.h b/moses/FF/CoveredReferenceFeature.h index e6fc79ff7..aedfb3793 100644 --- a/moses/FF/CoveredReferenceFeature.h +++ b/moses/FF/CoveredReferenceFeature.h @@ -53,10 +53,6 @@ public: return new CoveredReferenceState(); } - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const; void EvaluateWithSourceContext(const InputType &input , const InputPath &inputPath , const TargetPhrase &targetPhrase @@ -64,10 +60,6 @@ public: , ScoreComponentCollection &scoreBreakdown , ScoreComponentCollection *estimatedScores = NULL) const; - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/FF/DistortionScoreProducer.h b/moses/FF/DistortionScoreProducer.h index 6e6bfbaeb..cfe0dc005 100644 --- a/moses/FF/DistortionScoreProducer.h +++ b/moses/FF/DistortionScoreProducer.h @@ -47,23 +47,6 @@ public: throw std::logic_error("DistortionScoreProducer not supported in chart decoder, yet"); } - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const { - } }; } diff --git a/moses/FF/LexicalReordering/LexicalReordering.h b/moses/FF/LexicalReordering/LexicalReordering.h index c47e7f037..fa2747c82 100644 --- a/moses/FF/LexicalReordering/LexicalReordering.h +++ b/moses/FF/LexicalReordering/LexicalReordering.h @@ -64,28 +64,6 @@ public: UTIL_THROW2("LexicalReordering is not valid for chart decoder"); } - void - EvaluateWithSourceContext - (const InputType &input, - const InputPath &inputPath, - const TargetPhrase &targetPhrase, - const StackVec *stackVec, - ScoreComponentCollection& scoreBreakdown, - ScoreComponentCollection* estimatedScores = NULL) const - { } - - void - EvaluateTranslationOptionListWithSourceContext - (const InputType &input, const TranslationOptionList &transOptList) const - { } - - void - EvaluateInIsolation(const Phrase &source, - const TargetPhrase &targetPhrase, - ScoreComponentCollection &scoreBreakdown, - ScoreComponentCollection &estimatedScores) const - { } - bool GetHaveDefaultScores() { return m_haveDefaultScores; diff --git a/moses/FF/OSM-Feature/OpSequenceModel.h b/moses/FF/OSM-Feature/OpSequenceModel.h index b97450d6b..36a901974 100644 --- a/moses/FF/OSM-Feature/OpSequenceModel.h +++ b/moses/FF/OSM-Feature/OpSequenceModel.h @@ -37,18 +37,6 @@ public: int /* featureID - used to index the state in the previous hypotheses */, ScoreComponentCollection* accumulator) const; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - void EvaluateInIsolation(const Phrase &source , const TargetPhrase &targetPhrase , ScoreComponentCollection &scoreBreakdown diff --git a/moses/FF/PhraseBoundaryFeature.h b/moses/FF/PhraseBoundaryFeature.h index 916225b7d..9e84aaeef 100644 --- a/moses/FF/PhraseBoundaryFeature.h +++ b/moses/FF/PhraseBoundaryFeature.h @@ -54,23 +54,6 @@ public: throw std::logic_error("PhraseBoundaryState not supported in chart decoder, yet"); } - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const { - } - void SetParameter(const std::string& key, const std::string& value); private: diff --git a/moses/FF/PhraseOrientationFeature.h b/moses/FF/PhraseOrientationFeature.h index 82d3928d9..4d9dc20dc 100644 --- a/moses/FF/PhraseOrientationFeature.h +++ b/moses/FF/PhraseOrientationFeature.h @@ -296,18 +296,6 @@ public: , ScoreComponentCollection &scoreBreakdown , ScoreComponentCollection &estimatedScores) const; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const - {}; - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const - {} - FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/FF/SkeletonStatefulFF.cpp b/moses/FF/SkeletonStatefulFF.cpp index 4b077335e..2acaf2d2e 100644 --- a/moses/FF/SkeletonStatefulFF.cpp +++ b/moses/FF/SkeletonStatefulFF.cpp @@ -15,12 +15,19 @@ SkeletonStatefulFF::SkeletonStatefulFF(const std::string &line) ReadParameters(); } + +// An empty implementation of this function is provided by StatefulFeatureFunction. +// Unless you are actually implementing this, please remove it from your +// implementation (and the declaration in the header file to reduce code clutter. void SkeletonStatefulFF::EvaluateInIsolation(const Phrase &source , const TargetPhrase &targetPhrase , ScoreComponentCollection &scoreBreakdown , ScoreComponentCollection &estimatedScores) const {} +// An empty implementation of this function is provided by StatefulFeatureFunction. +// Unless you are actually implementing this, please remove it from your +// implementation (and the declaration in the header file to reduce code clutter. void SkeletonStatefulFF::EvaluateWithSourceContext(const InputType &input , const InputPath &inputPath , const TargetPhrase &targetPhrase @@ -29,8 +36,11 @@ void SkeletonStatefulFF::EvaluateWithSourceContext(const InputType &input , ScoreComponentCollection *estimatedScores) const {} -void SkeletonStatefulFF::EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const +// An empty implementation of this function is provided by StatefulFeatureFunction. +// Unless you are actually implementing this, please remove it from your +// implementation (and the declaration in the header file to reduce code clutter. +void SkeletonStatefulFF::EvaluateTranslationOptionListWithSourceContext +(const InputType &input, const TranslationOptionList &translationOptionList) const {} FFState* SkeletonStatefulFF::EvaluateWhenApplied( diff --git a/moses/FF/SkeletonStatefulFF.h b/moses/FF/SkeletonStatefulFF.h index bed54753c..7544ddd30 100644 --- a/moses/FF/SkeletonStatefulFF.h +++ b/moses/FF/SkeletonStatefulFF.h @@ -37,19 +37,35 @@ public: return new SkeletonState(0); } - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const; - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const; + // An empty implementation of this function is provided by StatefulFeatureFunction. + // Unless you are actually implementing this, please remove this declaration here + // and the empty skeleton implementation from the corresponding .cpp + // file to reduce code clutter. + void + EvaluateInIsolation(const Phrase &source + , const TargetPhrase &targetPhrase + , ScoreComponentCollection &scoreBreakdown + , ScoreComponentCollection &estimatedScores) const; + + // An empty implementation of this function is provided by StatefulFeatureFunction. + // Unless you are actually implementing this, please remove this declaration here + // and the empty skeleton implementation from the corresponding .cpp + // file to reduce code clutter. + void + EvaluateWithSourceContext(const InputType &input + , const InputPath &inputPath + , const TargetPhrase &targetPhrase + , const StackVec *stackVec + , ScoreComponentCollection &scoreBreakdown + , ScoreComponentCollection *estimatedScores = NULL) const; + + // An empty implementation of this function is provided by StatefulFeatureFunction. + // Unless you are actually implementing this, please remove this declaration here + // and the empty skeleton implementation from the corresponding .cpp + // file to reduce code clutter. + void + EvaluateTranslationOptionListWithSourceContext + ( const InputType &input , const TranslationOptionList &translationOptionList) const; FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, diff --git a/moses/FF/StatefulFeatureFunction.h b/moses/FF/StatefulFeatureFunction.h index ac6888aa7..5c8c2bbd6 100644 --- a/moses/FF/StatefulFeatureFunction.h +++ b/moses/FF/StatefulFeatureFunction.h @@ -70,6 +70,23 @@ public: return false; } + + virtual void + EvaluateInIsolation + (Phrase const& source, TargetPhrase const& targetPhrase, + ScoreComponentCollection &scoreBreakdown, + ScoreComponentCollection &estimatedScores) const {} + + virtual void + EvaluateWithSourceContext + (InputType const&input, InputPath const& inputPath, TargetPhrase const& targetPhrase, + StackVec const* stackVec, ScoreComponentCollection &scoreBreakdown, + ScoreComponentCollection *estimatedFutureScore = NULL) const {} + + virtual void + EvaluateTranslationOptionListWithSourceContext + (const InputType &input, const TranslationOptionList &translationOptionList) const {} + }; diff --git a/moses/FF/TargetBigramFeature.h b/moses/FF/TargetBigramFeature.h index 200eca060..eacd27656 100644 --- a/moses/FF/TargetBigramFeature.h +++ b/moses/FF/TargetBigramFeature.h @@ -48,22 +48,6 @@ public: ScoreComponentCollection* ) const { throw std::logic_error("TargetBigramFeature not valid in chart decoder"); } - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } void SetParameter(const std::string& key, const std::string& value); diff --git a/moses/FF/TargetNgramFeature.h b/moses/FF/TargetNgramFeature.h index 5bc2fc953..0a4b4aa25 100644 --- a/moses/FF/TargetNgramFeature.h +++ b/moses/FF/TargetNgramFeature.h @@ -215,24 +215,6 @@ public: virtual FFState* EvaluateWhenApplied(const ChartHypothesis& cur_hypo, int featureId, ScoreComponentCollection* accumulator) const; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const { - } - void SetParameter(const std::string& key, const std::string& value); private: diff --git a/moses/FF/TreeStructureFeature.h b/moses/FF/TreeStructureFeature.h index 361e8cc6a..353328466 100644 --- a/moses/FF/TreeStructureFeature.h +++ b/moses/FF/TreeStructureFeature.h @@ -63,22 +63,6 @@ public: void SetParameter(const std::string& key, const std::string& value); - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const {}; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const {}; - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - - FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/Hypothesis.h b/moses/Hypothesis.h index 01ba5a55e..cd26f5a2d 100644 --- a/moses/Hypothesis.h +++ b/moses/Hypothesis.h @@ -268,8 +268,8 @@ public: #ifdef HAVE_XMLRPC_C // these are implemented in moses/server/Hypothesis_4server.cpp ! - void OutputWordAlignment(std::vector<xmlrpc_c::value>& out, const ReportingOptions &options) const; - void OutputLocalWordAlignment(std::vector<xmlrpc_c::value>& dest, const ReportingOptions &options) const; + void OutputWordAlignment(std::vector<xmlrpc_c::value>& out) const; + void OutputLocalWordAlignment(std::vector<xmlrpc_c::value>& dest) const; #endif bool beats(Hypothesis const& b) const; diff --git a/moses/LM/Base.cpp b/moses/LM/Base.cpp index 06911c325..a89bd27c1 100644 --- a/moses/LM/Base.cpp +++ b/moses/LM/Base.cpp @@ -57,10 +57,11 @@ void LanguageModel::ReportHistoryOrder(std::ostream &out,const Phrase &phrase) c // out << "ReportHistoryOrder not implemented"; } -void LanguageModel::EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const +void +LanguageModel:: +EvaluateInIsolation(Phrase const& source, TargetPhrase const& targetPhrase, + ScoreComponentCollection &scoreBreakdown, + ScoreComponentCollection &estimatedScores) const { // contains factors used by this LM float fullScore, nGramScore; diff --git a/moses/LM/Base.h b/moses/LM/Base.h index 6212611ee..a052cc82e 100644 --- a/moses/LM/Base.h +++ b/moses/LM/Base.h @@ -90,18 +90,6 @@ public: , ScoreComponentCollection &scoreBreakdown , ScoreComponentCollection &estimatedScores) const; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores = NULL) const { - } - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const { - } - }; } diff --git a/moses/LM/BilingualLM.cpp b/moses/LM/BilingualLM.cpp index a52ccc1f2..b8974bb5e 100644 --- a/moses/LM/BilingualLM.cpp +++ b/moses/LM/BilingualLM.cpp @@ -188,22 +188,6 @@ size_t BilingualLM::getState(const Hypothesis& cur_hypo) const return hashCode; } -void BilingualLM::EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedScores) const {} - -void BilingualLM::EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedScores) const -{ - -} - - FFState* BilingualLM::EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/LM/BilingualLM.h b/moses/LM/BilingualLM.h index 5070c7cec..cb5075fd1 100644 --- a/moses/LM/BilingualLM.h +++ b/moses/LM/BilingualLM.h @@ -119,23 +119,6 @@ public: void Load(); - void EvaluateInIsolation( - const Phrase &source, - const TargetPhrase &targetPhrase, - ScoreComponentCollection &scoreBreakdown, - ScoreComponentCollection &estimatedScores) const; - - void EvaluateWithSourceContext( - const InputType &input, - const InputPath &inputPath, - const TargetPhrase &targetPhrase, - const StackVec *stackVec, - ScoreComponentCollection &scoreBreakdown, - ScoreComponentCollection *estimatedScores = NULL) const; - - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const {}; - FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/LM/RDLM.h b/moses/LM/RDLM.h index 8e169310f..963c1e8d5 100644 --- a/moses/LM/RDLM.h +++ b/moses/LM/RDLM.h @@ -196,18 +196,7 @@ public: } void SetParameter(const std::string& key, const std::string& value); - void EvaluateInIsolation(const Phrase &source - , const TargetPhrase &targetPhrase - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection &estimatedFutureScore) const {}; - void EvaluateWithSourceContext(const InputType &input - , const InputPath &inputPath - , const TargetPhrase &targetPhrase - , const StackVec *stackVec - , ScoreComponentCollection &scoreBreakdown - , ScoreComponentCollection *estimatedFutureScore = NULL) const {}; - void EvaluateTranslationOptionListWithSourceContext(const InputType &input - , const TranslationOptionList &translationOptionList) const {}; + FFState* EvaluateWhenApplied( const Hypothesis& cur_hypo, const FFState* prev_state, diff --git a/moses/ThreadPool.cpp b/moses/ThreadPool.cpp index cba459697..deb6acb67 100644 --- a/moses/ThreadPool.cpp +++ b/moses/ThreadPool.cpp @@ -55,9 +55,6 @@ void ThreadPool::Execute() } //Execute job if (task) { - // must read from task before run. otherwise task may be deleted by main thread - // race condition - task->DeleteAfterExecution(); task->Run(); } m_threadAvailable.notify_all(); diff --git a/moses/ThreadPool.h b/moses/ThreadPool.h index b7d459bb2..f5be5e52b 100644 --- a/moses/ThreadPool.h +++ b/moses/ThreadPool.h @@ -53,9 +53,6 @@ class Task { public: virtual void Run() = 0; - virtual bool DeleteAfterExecution() { - return true; - } virtual ~Task() {} }; diff --git a/moses/TranslationModel/UG/bitext-find.cc b/moses/TranslationModel/UG/bitext-find.cc index d02ce0710..4c6e13208 100644 --- a/moses/TranslationModel/UG/bitext-find.cc +++ b/moses/TranslationModel/UG/bitext-find.cc @@ -53,16 +53,16 @@ int main(int argc, char* argv[]) interpret_args(argc, argv); if (Q1.empty() && Q2.empty()) exit(0); - mmbitext B; string w; - B.open(bname, L1, L2); + boost::shared_ptr<mmbitext> B(new mmbitext); string w; + B->open(bname, L1, L2); - Bitext<Token>::iter m1(B.I1.get(), *B.V1, Q1); + Bitext<Token>::iter m1(B->I1.get(), *B->V1, Q1); if (Q1.size() && m1.size() == 0) exit(0); - Bitext<Token>::iter m2(B.I2.get(), *B.V2, Q2); + Bitext<Token>::iter m2(B->I2.get(), *B->V2, Q2); if (Q2.size() && m2.size() == 0) exit(0); - bitvector check(B.T1->size()); + bitvector check(B->T1->size()); if (Q1.size() == 0 || Q2.size() == 0) check.set(); else (m2.markSentences(check)); @@ -87,23 +87,23 @@ int main(int argc, char* argv[]) size_t s1,s2,e1,e2; int po_fwd=-1,po_bwd=-1; std::vector<unsigned char> caln; - // cout << sid << " " << B.docname(sid) << std::endl; - if (!B.find_trg_phr_bounds(sid, off, off+m.size(), + // cout << sid << " " << B->docname(sid) << std::endl; + if (!B->find_trg_phr_bounds(sid, off, off+m.size(), s1,s2,e1,e2,po_fwd,po_bwd, &caln, NULL, &m == &m2)) { // cout << "alignment failure" << std::endl; } - std::cout << sid << " " << B.docname(sid) + std::cout << sid << " " << B->sid2docname(sid) << " dfwd=" << po_fwd << " dbwd=" << po_bwd << "\n"; - write_sentence(*B.T1, sid, *B.V1, std::cout); std::cout << "\n"; - write_sentence(*B.T2, sid, *B.V2, std::cout); std::cout << "\n"; - B.write_yawat_alignment(sid, - m1.size() ? &m1 : NULL, - m2.size() ? &m2 : NULL, std::cout); + write_sentence(*B->T1, sid, *B->V1, std::cout); std::cout << "\n"; + write_sentence(*B->T2, sid, *B->V2, std::cout); std::cout << "\n"; + B->write_yawat_alignment(sid, + m1.size() ? &m1 : NULL, + m2.size() ? &m2 : NULL, std::cout); std::cout << std::endl; } diff --git a/moses/TranslationModel/UG/check-coverage.cc b/moses/TranslationModel/UG/check-coverage.cc index 28d3e8968..3cf954912 100644 --- a/moses/TranslationModel/UG/check-coverage.cc +++ b/moses/TranslationModel/UG/check-coverage.cc @@ -41,8 +41,8 @@ basename(string const path, string const suffix) int main(int argc, char* argv[]) { - bitext_t B; - B.open(argv[1],argv[2],argv[3]); + boost::shared_ptr<bitext_t> B(new bitext_t); + B->open(argv[1],argv[2],argv[3]); string line; string ifile = argv[4]; string docname = basename(ifile, string(".") + argv[2] + ".gz"); @@ -52,10 +52,10 @@ int main(int argc, char* argv[]) { cout << line << " [" << docname << "]" << endl; vector<id_type> snt; - B.V1->fillIdSeq(line,snt); + B->V1->fillIdSeq(line,snt); for (size_t i = 0; i < snt.size(); ++i) { - bitext_t::iter m(B.I1.get()); + bitext_t::iter m(B->I1.get()); for (size_t k = i; k < snt.size() && m.extend(snt[k]); ++k) { if (m.ca() > 500) continue; @@ -65,9 +65,10 @@ int main(int argc, char* argv[]) while (I.next != stop) { m.root->readEntry(I.next,I); - ++cnt[B.docname(I.sid)]; + ++cnt[B->sid2docname(I.sid)]; } - cout << setw(8) << int(m.ca()) << " " << B.V1->toString(&snt[i],&snt[k+1]) << endl; + cout << setw(8) << int(m.ca()) << " " + << B->V1->toString(&snt[i],&snt[k+1]) << endl; typedef pair<string,uint32_t> entry; vector<entry> ranked; ranked.reserve(cnt.size()); BOOST_FOREACH(entry const& e, cnt) ranked.push_back(e); diff --git a/moses/TranslationModel/UG/check-coverage3.cc b/moses/TranslationModel/UG/check-coverage3.cc index d41e10ef3..c4a83877f 100644 --- a/moses/TranslationModel/UG/check-coverage3.cc +++ b/moses/TranslationModel/UG/check-coverage3.cc @@ -37,7 +37,7 @@ basename(string const path, string const suffix) size_t k = path.size() - suffix.size(); cout << path << " " << suffix << endl; cout << path.substr(0,p) << " " << path.substr(k) << endl; - return path.substr(p, suffix == &path[k] ? k-p : path.size() - p); + return path.substr(p+1, suffix == &path[k] ? k-p-1 : path.size() - p); } int main(int argc, char* argv[]) @@ -47,6 +47,7 @@ int main(int argc, char* argv[]) string line; string ifile = argv[4]; string docname = basename(ifile, string(".") + argv[2] + ".gz"); + id_type docid = B->docname2docid(docname); boost::iostreams::filtering_istream in; ugdiss::open_input_stream(ifile,in); while(getline(in,line)) @@ -57,13 +58,70 @@ int main(int argc, char* argv[]) for (size_t i = 0; i < snt.size(); ++i) { bitext_t::iter m(B->I1.get()); - for (size_t k = i; k < snt.size() && m.extend(snt[k]); ++k) + for (size_t k = i; k < snt.size() && m.extend(snt[k]); ++k); + for (size_t num_occurrences = m.ca(); m.size(); m.up()) { + if (size_t(m.ca()) == num_occurrences) continue; + num_occurrences = m.ca(); SPTR<SamplingBias const> zilch; BitextSampler<Token> s(B.get(), m, zilch, 1000, 1000, sapt::random_sampling); s(); - cout << m.size() << " " << s.stats()->trg.size() << endl; + if (s.stats()->trg.size() == 0) continue; + // if (s.stats()->indoc[docname] > 10) continue; + sapt::pstats::indoc_map_t::const_iterator d + = s.stats()->indoc.find(docid); + size_t indoccnt = d != s.stats()->indoc.end() ? d->second : 0; + cout << m.size() << " : " << m.str(B->V1.get()) << " (" + << s.stats()->trg.size() << " entries; " + << indoccnt << "/" << s.stats()->good + << " samples in domain)" << endl; + vector<PhrasePair<Token> > ppairs; + PhrasePair<Token>::SortDescendingByJointCount sorter; + expand(m,*B,*s.stats(),ppairs,NULL); + sort(ppairs.begin(),ppairs.end(),sorter); + boost::format fmt("%4d/%d/%d |%s| (%4.2f : %4.2f)"); + BOOST_FOREACH(PhrasePair<Token>& ppair, ppairs) + { + if (ppair.joint * 100 < ppair.good1) break; + ppair.good2 = ppair.raw2 * float(ppair.good1)/ppair.raw1; + ppair.good2 = max(ppair.good2, ppair.joint); + +#if 0 + cout << "\t" + << (fmt % ppair.joint % ppair.good1 % ppair.good2 + % B->T2->pid2str(B->V2.get(),ppair.p2) + % (float(ppair.joint)/ppair.good1) + % (float(ppair.joint)/ppair.good2) + ) << "\n"; + typedef std::map<uint32_t, uint32_t>::const_iterator iter; + for (iter d = ppair.indoc.begin(); d != ppair.indoc.end(); ++d) + { + // if (d != ppair.indoc.begin()) cout << "; "; + cout << (boost::format("\t\t%4d %s") % d->second + % B->docid2name(d->first)) + << endl; + } + cout << endl; +#else + cout << "\t" + << (fmt % ppair.joint % ppair.good1 % ppair.good2 + % B->T2->pid2str(B->V2.get(),ppair.p2) + % (float(ppair.joint)/ppair.good1) + % (float(ppair.joint)/ppair.good2) + ) << " ["; + typedef std::map<uint32_t, uint32_t>::const_iterator iter; + for (iter d = ppair.indoc.begin(); d != ppair.indoc.end(); ++d) + { + if (d != ppair.indoc.begin()) cout << "; "; + cout << (boost::format("%s: %d") % B->docid2name(d->first) + % d->second) ; + } + cout << "]" << endl; + +#endif + + } } } } diff --git a/moses/TranslationModel/UG/mm/ug_bitext.h b/moses/TranslationModel/UG/mm/ug_bitext.h index aeb8703dc..4094dc9cc 100644 --- a/moses/TranslationModel/UG/mm/ug_bitext.h +++ b/moses/TranslationModel/UG/mm/ug_bitext.h @@ -217,17 +217,42 @@ namespace sapt write_yawat_alignment ( id_type const sid, iter const* m1, iter const* m2, std::ostream& out ) const; - std::string docname(id_type const sid) const; - + std::string sid2docname(id_type const sid) const; + std::string docid2name(id_type const sid) const; + int docname2docid(std::string const& name) const; + std::vector<id_type> const* sid2did() const; + int sid2did(uint32_t sid) const; }; #include "ug_bitext_agenda.h" template<typename Token> + int + Bitext<Token>:: + docname2docid(std::string const& name) const + { + std::map<std::string,id_type>::const_iterator m; + m = m_docname2docid.find(name); + if (m != m_docname2docid.end()) return m->second; + return -1; + } + + template<typename Token> std::string Bitext<Token>:: - docname(id_type const sid) const + docid2name(id_type const did) const + { + if (did < m_docname.size()) + return m_docname[did]; + else + return (boost::format("%d") % did).str(); + } + + template<typename Token> + std::string + Bitext<Token>:: + sid2docname(id_type const sid) const { if (sid < m_sid2docid->size() && (*m_sid2docid)[sid] < m_docname.size()) return m_docname[(*m_sid2docid)[sid]]; @@ -244,6 +269,17 @@ namespace sapt } template<typename Token> + int + Bitext<Token>:: + sid2did(uint32_t sid) const + { + if (m_sid2docid) + return m_sid2docid->at(sid); + return -1; + } + + + template<typename Token> SPTR<SentenceBias> Bitext<Token>:: loadSentenceBias(std::string const& fname) const diff --git a/moses/TranslationModel/UG/mm/ug_bitext_sampler.h b/moses/TranslationModel/UG/mm/ug_bitext_sampler.h index a36e0772f..a10c83c6e 100644 --- a/moses/TranslationModel/UG/mm/ug_bitext_sampler.h +++ b/moses/TranslationModel/UG/mm/ug_bitext_sampler.h @@ -52,7 +52,7 @@ BitextSampler : public Moses::reference_counter // const members // SPTR<bitext const> const m_bitext; // keep bitext alive while I am // should be an - iptr<bitext const> const m_bitext; // keep bitext alive as long as I am + SPTR<bitext const> const m_bitext; // keep bitext alive as long as I am size_t const m_plen; // length of lookup phrase bool const m_fwd; // forward or backward direction? SPTR<tsa const> const m_root; // root of suffix array @@ -275,7 +275,7 @@ consider_sample(TokenPosition const& p) bitvector full_aln(100*100); PhraseExtractionRecord rec(p.sid, p.offset, p.offset + m_plen, !m_fwd, &aln, &full_aln); - int docid = m_bias ? m_bias->GetClass(p.sid) : -1; + int docid = m_bias ? m_bias->GetClass(p.sid) : m_bitext->sid2did(p.sid); if (!m_bitext->find_trg_phr_bounds(rec)) { // no good, probably because phrase is not coherent m_stats->count_sample(docid, 0, rec.po_fwd, rec.po_bwd); diff --git a/moses/TranslationModel/UG/mm/ug_prep_phrases.h b/moses/TranslationModel/UG/mm/ug_prep_phrases.h index 1c62db2e3..1d04d6680 100644 --- a/moses/TranslationModel/UG/mm/ug_prep_phrases.h +++ b/moses/TranslationModel/UG/mm/ug_prep_phrases.h @@ -16,7 +16,7 @@ struct StatsCollector typedef lru_cache::LRU_Cache< uint64_t, pstats > hcache_t; typedef ThreadSafeContainer<uint64_t, SPTR<pstats> > pcache_t; typedef map<uint64_t, SPTR<pstats> > lcache_t; - iptr<Bitext<Token> const> bitext; // underlying bitext + SPTR<Bitext<Token> const> bitext; // underlying bitext sampling_method method; // sampling method size_t sample_size; // sample size SPTR<SamplingBias const> bias; // sampling bias @@ -26,7 +26,7 @@ struct StatsCollector SPTR<lcache_t> lcache; // local cache ug::ThreadPool* tpool; // thread pool to run jobs on - StatsCollector(iptr<Bitext<Token> > xbitext, + StatsCollector(SPTR<Bitext<Token> > xbitext, SPTR<SamplingBias> const xbias) : method(ranked_sampling) , sample_size(100) diff --git a/moses/TranslationModel/UG/mmsapt.h b/moses/TranslationModel/UG/mmsapt.h index d30177518..9dbd869df 100644 --- a/moses/TranslationModel/UG/mmsapt.h +++ b/moses/TranslationModel/UG/mmsapt.h @@ -71,7 +71,7 @@ namespace Moses typedef sapt::PhraseScorer<Token> pscorer; private: // vector<SPTR<bitext> > shards; - iptr<mmbitext> btfix; + SPTR<mmbitext> btfix; SPTR<imbitext> btdyn; std::string m_bname, m_extra_data, m_bias_file,m_bias_server; std::string L1; @@ -160,7 +160,7 @@ namespace Moses #if PROVIDES_RANKED_SAMPLING void - set_bias_for_ranking(ttasksptr const& ttask, iptr<sapt::Bitext<Token> const> bt); + set_bias_for_ranking(ttasksptr const& ttask, SPTR<sapt::Bitext<Token> const> bt); #endif private: diff --git a/moses/parameters/AllOptions.cpp b/moses/parameters/AllOptions.cpp index cfb925dfb..a7acdadfa 100644 --- a/moses/parameters/AllOptions.cpp +++ b/moses/parameters/AllOptions.cpp @@ -65,6 +65,7 @@ namespace Moses // set m_nbest_options.enabled = true if necessary: nbest.enabled = (nbest.enabled || mira || search.consensus || nbest.nbest_size > 0 + || mbr.enabled || lmbr.enabled || !output.SearchGraph.empty() || !output.SearchGraphExtended.empty() || !output.SearchGraphSLF.empty() diff --git a/moses/parameters/NBestOptions.cpp b/moses/parameters/NBestOptions.cpp index e916c3437..c65f9e852 100644 --- a/moses/parameters/NBestOptions.cpp +++ b/moses/parameters/NBestOptions.cpp @@ -43,7 +43,7 @@ update(std::map<std::string,xmlrpc_c::value>const& param) params_t::const_iterator si = param.find("nbest"); if (si != param.end()) nbest_size = xmlrpc_c::value_int(si->second); - only_distinct = check(param, "nbest-distinct"); + only_distinct = check(param, "nbest-distinct", only_distinct); enabled = (nbest_size > 0); return true; } diff --git a/moses/parameters/OptionsBaseClass.cpp b/moses/parameters/OptionsBaseClass.cpp index 148aa5d24..a19aaf7a6 100644 --- a/moses/parameters/OptionsBaseClass.cpp +++ b/moses/parameters/OptionsBaseClass.cpp @@ -1,5 +1,7 @@ // -*- mode: c++; indent-tabs-mode: nil; tab-width:2 -*- #include "OptionsBaseClass.h" +#include "moses/Util.h" + namespace Moses { #ifdef HAVE_XMLRPC_C @@ -15,10 +17,12 @@ namespace Moses { bool OptionsBaseClass:: check(std::map<std::string, xmlrpc_c::value> const& param, - std::string const key) + std::string const key, bool dfltval) { std::map<std::string, xmlrpc_c::value>::const_iterator m; - return (param.find(key) != param.end()); + m = param.find(key); + if (m == param.end()) return dfltval; + return Scan<bool>(xmlrpc_c::value_string(m->second)); } #endif } diff --git a/moses/parameters/OptionsBaseClass.h b/moses/parameters/OptionsBaseClass.h index 71f9fa77a..ddec48192 100644 --- a/moses/parameters/OptionsBaseClass.h +++ b/moses/parameters/OptionsBaseClass.h @@ -13,6 +13,6 @@ namespace Moses #endif bool check(std::map<std::string, xmlrpc_c::value> const& param, - std::string const key); + std::string const key, bool dfltval); }; } diff --git a/moses/parameters/ReportingOptions.cpp b/moses/parameters/ReportingOptions.cpp index 640b6a177..10a0cfcb4 100644 --- a/moses/parameters/ReportingOptions.cpp +++ b/moses/parameters/ReportingOptions.cpp @@ -75,7 +75,7 @@ namespace Moses { ReportingOptions:: update(std::map<std::string, xmlrpc_c::value>const& param) { - ReportAllFactors = check(param, "report-all-factors"); + ReportAllFactors = check(param, "report-all-factors", ReportAllFactors); return true; } #endif diff --git a/moses/server/Hypothesis_4server.cpp b/moses/server/Hypothesis_4server.cpp index a7049f3d3..372755750 100644 --- a/moses/server/Hypothesis_4server.cpp +++ b/moses/server/Hypothesis_4server.cpp @@ -5,14 +5,15 @@ namespace Moses { void Hypothesis:: - OutputLocalWordAlignment(std::vector<xmlrpc_c::value>& dest, const ReportingOptions &options) const + OutputLocalWordAlignment(std::vector<xmlrpc_c::value>& dest) const { using namespace std; Range const& src = this->GetCurrSourceWordsRange(); Range const& trg = this->GetCurrTargetWordsRange(); + WordAlignmentSort waso = m_manager.options().output.WA_SortOrder; vector<pair<size_t,size_t> const* > a - = this->GetCurrTargetPhrase().GetAlignTerm().GetSortedAlignments(options.WA_SortOrder); + = this->GetCurrTargetPhrase().GetAlignTerm().GetSortedAlignments(waso); typedef pair<size_t,size_t> item; map<string, xmlrpc_c::value> M; BOOST_FOREACH(item const* p, a) { @@ -24,13 +25,13 @@ namespace Moses { void Hypothesis:: - OutputWordAlignment(std::vector<xmlrpc_c::value>& out, const ReportingOptions &options) const + OutputWordAlignment(std::vector<xmlrpc_c::value>& out) const { std::vector<Hypothesis const*> tmp; for (Hypothesis const* h = this; h; h = h->GetPrevHypo()) tmp.push_back(h); for (size_t i = tmp.size(); i-- > 0;) - tmp[i]->OutputLocalWordAlignment(out, options); + tmp[i]->OutputLocalWordAlignment(out); } } diff --git a/moses/server/TranslationRequest.cpp b/moses/server/TranslationRequest.cpp index 1748f80ae..2157b2466 100644 --- a/moses/server/TranslationRequest.cpp +++ b/moses/server/TranslationRequest.cpp @@ -3,6 +3,8 @@ #include "moses/ContextScope.h" #include <boost/foreach.hpp> #include "moses/Util.h" +#include "moses/Hypothesis.h" + namespace MosesServer { using namespace std; @@ -164,15 +166,15 @@ insertGraphInfo(Manager& manager, map<string, xmlrpc_c::value>& retData) retData["sg"] = xmlrpc_c::value_array(searchGraphXml); } -void -TranslationRequest:: -output_phrase(ostream& out, Phrase const& phrase) const -{ - if (!m_options.output.ReportAllFactors) { - for (size_t i = 0 ; i < phrase.GetSize(); ++i) - out << *phrase.GetFactor(i, 0) << " "; - } else out << phrase; -} +// void +// TranslationRequest:: +// output_phrase(ostream& out, Phrase const& phrase) const +// { +// if (!m_options.output.ReportAllFactors) { +// for (size_t i = 0 ; i < phrase.GetSize(); ++i) +// out << *phrase.GetFactor(i, 0) << " "; +// } else out << phrase; +// } void TranslationRequest:: @@ -193,7 +195,7 @@ outputNBest(const Manager& manager, map<string, xmlrpc_c::value>& retData) vector<const Hypothesis *> const& E = path->GetEdges(); if (!E.size()) continue; std::map<std::string, xmlrpc_c::value> nBestXmlItem; - pack_hypothesis(E, "hyp", nBestXmlItem); + pack_hypothesis(manager, E, "hyp", nBestXmlItem); if (m_withScoreBreakdown) { // should the score breakdown be reported in a more structured manner? ostringstream buf; @@ -262,8 +264,11 @@ bool check(std::map<std::string, xmlrpc_c::value> const& param, std::string const key) { - std::map<std::string, xmlrpc_c::value>::const_iterator m; - return (param.find(key) != param.end()); + std::map<std::string, xmlrpc_c::value>::const_iterator m = param.find(key); + if(m == param.end()) return false; + std::string val = string(xmlrpc_c::value_string(m->second)); + if(val == "true" || val == "True" || val == "TRUE" || val == "1") return true; + return false; } void @@ -367,15 +372,17 @@ run_chart_decoder() void TranslationRequest:: -pack_hypothesis(vector<Hypothesis const* > const& edges, string const& key, +pack_hypothesis(const Moses::Manager& manager, vector<Hypothesis const* > const& edges, string const& key, map<string, xmlrpc_c::value> & dest) const { // target string ostringstream target; - BOOST_REVERSE_FOREACH(Hypothesis const* e, edges) - output_phrase(target, e->GetCurrTargetPhrase()); - XVERBOSE(1,"SERVER TRANSLATION: " << target.str() << std::endl); - + BOOST_REVERSE_FOREACH(Hypothesis const* e, edges) { + manager.OutputSurface(target, *e, m_options.output.factor_order, + m_options.output.ReportSegmentation, m_options.output.ReportAllFactors); + } + XVERBOSE(1, "BEST TRANSLATION: " << *(manager.GetBestHypothesis()) << std::endl); +// XVERBOSE(1,"SERVER TRANSLATION: " << target.str() << std::endl); dest[key] = xmlrpc_c::value_string(target.str()); if (m_withAlignInfo) { @@ -383,7 +390,7 @@ pack_hypothesis(vector<Hypothesis const* > const& edges, string const& key, vector<xmlrpc_c::value> p_aln; BOOST_REVERSE_FOREACH(Hypothesis const* e, edges) - add_phrase_aln_info(*e, p_aln); + add_phrase_aln_info(*e, p_aln); dest["align"] = xmlrpc_c::value_array(p_aln); } @@ -391,21 +398,21 @@ pack_hypothesis(vector<Hypothesis const* > const& edges, string const& key, // word alignment, if requested vector<xmlrpc_c::value> w_aln; BOOST_FOREACH(Hypothesis const* e, edges) - e->OutputLocalWordAlignment(w_aln, m_options.output); + e->OutputLocalWordAlignment(w_aln); dest["word-align"] = xmlrpc_c::value_array(w_aln); } } void TranslationRequest:: -pack_hypothesis(Hypothesis const* h, string const& key, +pack_hypothesis(const Moses::Manager& manager, Hypothesis const* h, string const& key, map<string, xmlrpc_c::value>& dest) const { using namespace std; vector<Hypothesis const*> edges; for (; h; h = h->GetPrevHypo()) edges.push_back(h); - pack_hypothesis(edges, key, dest); + pack_hypothesis(manager, edges, key, dest); } @@ -422,7 +429,7 @@ run_phrase_decoder() manager.Decode(); - pack_hypothesis(manager.GetBestHypothesis(), "text", m_retData); + pack_hypothesis(manager, manager.GetBestHypothesis(), "text", m_retData); if (m_session_id) m_retData["session-id"] = xmlrpc_c::value_int(m_session_id); diff --git a/moses/server/TranslationRequest.h b/moses/server/TranslationRequest.h index b93043b9b..7fb122714 100644 --- a/moses/server/TranslationRequest.h +++ b/moses/server/TranslationRequest.h @@ -58,17 +58,16 @@ TranslationRequest : public virtual Moses::TranslationTask run_phrase_decoder(); void - pack_hypothesis(std::vector<Moses::Hypothesis const* > const& edges, + pack_hypothesis(const Moses::Manager& manager, std::vector<Moses::Hypothesis const* > const& edges, std::string const& key, std::map<std::string, xmlrpc_c::value> & dest) const; void - pack_hypothesis(Moses::Hypothesis const* h, std::string const& key, + pack_hypothesis(const Moses::Manager& manager, Moses::Hypothesis const* h, std::string const& key, std::map<std::string, xmlrpc_c::value> & dest) const; - - void - output_phrase(std::ostream& out, Moses::Phrase const& phrase) const; + // void + // output_phrase(std::ostream& out, Moses::Phrase const& phrase) const; void add_phrase_aln_info(Moses::Hypothesis const& h, diff --git a/regression-testing/Jamfile b/regression-testing/Jamfile index 3f07744c6..2452b27a5 100644 --- a/regression-testing/Jamfile +++ b/regression-testing/Jamfile @@ -1,6 +1,7 @@ import option path ; with-regtest = [ option.get "with-regtest" ] ; +with-xmlrpc = [ option.get "with-xmlrpc-c" ] ; if $(with-regtest) { with-regtest = [ path.root $(with-regtest) [ path.pwd ] ] ; @@ -24,9 +25,20 @@ if $(with-regtest) { actions reg_test_decode { $(TOP)/regression-testing/run-single-test.perl --decoder=$(>) --test=$(<:B) --data-dir=$(with-regtest) --test-dir=$(test-dir) && touch $(<) } - reg_test phrase : [ glob $(test-dir)/phrase.* ] : ../moses-cmd//moses : @reg_test_decode ; - reg_test chart : [ glob $(test-dir)/chart.* ] : ../moses-cmd//moses : @reg_test_decode ; - + + if $(with-xmlrpc) { + actions reg_test_decode_server { + $(TOP)/regression-testing/run-single-test.perl --server --decoder=$(>) --test=$(<:B) --data-dir=$(with-regtest) --test-dir=$(test-dir) && touch $(<) + } + reg_test phrase-server : [ glob $(test-dir)/phrase-server.* ] : ../moses-cmd//moses : @reg_test_decode_server ; + } + reg_test phrase : [ glob $(test-dir)/phrase.* : $(test-dir)/*withDALM ] : ../moses-cmd//moses : @reg_test_decode ; + reg_test chart : [ glob $(test-dir)/chart.* : $(test-dir)/*withDALM ] : ../moses-cmd//moses : @reg_test_decode ; + if [ option.get "with-dalm" : : "yes" ] { + reg_test dalm : [ glob $(test-dir)/*withDALM ] : ../moses-cmd//moses : @reg_test_decode ; + } else { + alias dalm ; + } actions reg_test_score { $(TOP)/regression-testing/run-test-scorer.perl --scorer=$(>) --test=$(<:B) --data-dir=$(with-regtest) --test-dir=$(test-dir) && touch $(<) } @@ -55,5 +67,5 @@ if $(with-regtest) { reg_test misc : [ glob $(test-dir)/misc.* : $(test-dir)/misc.mml* ] : ..//prefix-bin ..//prefix-lib : @reg_test_misc ; reg_test misc-mml : [ glob $(test-dir)/misc.mml* ] : $(TOP)/scripts/ems/support/mml-filter.py $(TOP)/scripts/ems/support/defaultconfig.py : @reg_test_misc ; - alias all : phrase chart mert score extract extractrules misc misc-mml ; + alias all : phrase chart mert score extract extractrules misc misc-mml dalm ; } diff --git a/regression-testing/run-single-test.perl b/regression-testing/run-single-test.perl index 94a247e46..c99165247 100755 --- a/regression-testing/run-single-test.perl +++ b/regression-testing/run-single-test.perl @@ -2,6 +2,8 @@ # $Id$ +use Encode; +use utf8; use warnings; use strict; my $script_dir; BEGIN { use Cwd qw/ abs_path /; use File::Basename; $script_dir = dirname(abs_path($0)); push @INC, $script_dir; } @@ -9,6 +11,7 @@ use MosesRegressionTesting; use Getopt::Long; use File::Temp qw ( tempfile ); use POSIX qw ( strftime ); +use POSIX ":sys_wait_h"; my @SIGS = qw ( SIGHUP SIGINT SIGQUIT SIGILL SIGTRAP SIGABRT SIGIOT SIGBUS SIGFPE SIGKILL SIGUSR1 SIGSEGV SIGUSR2 SIGPIPE SIGALRM SIGTERM SIGSTKFLT SIGCHLD SIGCONT SIGSTOP SIGTSTP SIGTTIN SIGTTOU SIGURG SIGXCPU SIGXFSZ SIGVTALRM SIGPROF SIGWINCH SIGIO SIGPWR SIGSYS SIGUNUSED SIGRTMIN ); my ($decoder, $test_name); @@ -17,14 +20,31 @@ my $data_dir; my $BIN_TEST = $script_dir; my $results_dir; my $NBEST = 0; - +my $run_server_test = 0; +my $serverport = int(rand(9999)) + 10001; +my $url = "http://localhost:$serverport/RPC2"; +my $startupTest = 0; GetOptions("decoder=s" => \$decoder, "test=s" => \$test_name, "data-dir=s"=> \$data_dir, "test-dir=s"=> \$test_dir, "results-dir=s"=> \$results_dir, + "server"=> \$run_server_test, + "startuptest"=> \$startupTest ) or exit 1; +if($run_server_test) +{ + eval { + require XMLRPC::Lite; + import XMLRPC::Lite; + }; + if ($@) { + die "Error: XMLRPC::Lite not installed, moses server regression tests will not be run. $@"; + } + exit(0) if($startupTest); +} + die "Please specify a decoder with --decoder\n" unless $decoder; die "Please specify a test to run with --test\n" unless $test_name; @@ -72,8 +92,13 @@ if (!-d $truth) { } print "RESULTS AVAILABLE IN: $results\n\n"; - -my ($o, $elapsed, $ec, $sig) = exec_moses($decoder, $local_moses_ini, $input, $results); +my ($o, $elapsed, $ec, $sig); +if($run_server_test) { + ($o, $elapsed, $ec, $sig) = exec_moses_server($decoder, $local_moses_ini, $input, $results); +} +else { + ($o, $elapsed, $ec, $sig) = exec_moses($decoder, $local_moses_ini, $input, $results); +} my $error = ($sig || $ec > 0); if ($error) { open OUT, ">$results/Summary"; @@ -139,6 +164,54 @@ sub exec_moses { return ($o, $elapsed, $ec, $sig); } +sub exec_moses_server { + my ($decoder, $conf, $input, $results) = @_; + my $start_time = time; + my ($o, $ec, $sig); + $ec = 0; $sig = 0; $o = 0; + my $pid = fork(); + if (not defined $pid) { + warn "resources not avilable to fork Moses server\n"; + $ec = 1; # to generate error + } elsif ($pid == 0) { + setpgrp(0, 0); + warn "Starting Moses server on port $serverport ...\n"; + ($o, $ec, $sig) = run_command("$decoder --server --server-port $serverport -f $conf -verbose 2 --server-log $results/run.stderr.server 2> $results/run.stderr "); + exit; + # this should not be reached unless the server fails to start + } + while( 1==1 ) # wait until the server is listening for requests + { + sleep 5; + my $str = `grep "Listening on port $serverport" $results/run.stderr`; + last if($str =~ /Listening/); + } + my $proxy = XMLRPC::Lite->proxy($url); + warn "Opening file $input to write to $results\n"; + open(TEXTIN, "$input") or die "Can not open the input file to translate with Moses server\n"; + binmode TEXTIN, ':utf8'; + open(TEXTOUT, ">$results/run.stdout"); + binmode TEXTOUT, ':utf8'; + while(<TEXTIN>) + { + chop; + my $encoded = SOAP::Data->type(string => $_); # NOTE: assuming properly encoded UTF-8 input: check tests before adding them! + my %param = ("text" => $encoded); + my $result = $proxy->call("translate",\%param)->result; + print TEXTOUT $result->{'text'} . "\n"; + } + close(TEXTIN); + close(TEXTOUT); + my $elapsed = time - $start_time; + print STDERR "Finished translating file $input\n"; + if(waitpid($pid, WNOHANG) <= 0) + { + warn "Killing process group $pid of the $decoder --server ... \n"; + kill 9, -$pid; + } + return ($o, $elapsed, $ec, $sig); +} + sub run_command { my ($cmd) = @_; my $o = `$cmd`; diff --git a/regtest b/regtest -Subproject e07a00c9733e0fecb8433f1c9d5805d3f0b35c6 +Subproject 37a595fd7bf41226933c0fdb6fb792bdc877c3f diff --git a/run-regtests.sh b/run-regtests.sh index 8b64eac22..db4290a90 100755 --- a/run-regtests.sh +++ b/run-regtests.sh @@ -2,13 +2,21 @@ # this script assumes that all 3rd-party dependencies are installed under ./opt # you can install all 3rd-party dependencies by running make -f contrib/Makefiles/install-dependencies.gmake -set -e -o pipefail +set -e -o pipefail -x +opt=$(pwd)/opt git submodule init git submodule update regtest +if [ "$RECOMPILE" == "NO" ] ; then + RECOMPILE= +else + RECOMPILE="-a" +fi # test compilation without xmlrpc-c -./bjam -j$(nproc) --with-irstlm=./opt --with-boost=./opt --with-cmph=./opt --no-xmlrpc-c --with-regtest=./regtest -a -q $@ || exit $? +./bjam -j$(nproc) --with-irstlm=$opt --with-boost=$opt --with-cmph=$opt --no-xmlrpc-c --with-regtest=$(pwd)/regtest -a -q $@ || exit $? # test compilation with xmlrpc-c -./bjam -j$(nproc) --with-irstlm=./opt --with-boost=./opt --with-cmph=./opt --with-xmlrpc-c=./opt --with-regtest=./regtest -a -q $@ +if ./regression-testing/run-single-test.perl --server --startuptest ; then + ./bjam -j$(nproc) --with-irstlm=$opt --with-boost=$opt --with-cmph=$opt --with-xmlrpc-c=$opt --with-regtest=$(pwd)/regtest $RECOMPILE -q $@ +fi diff --git a/scripts/ems/experiment.meta b/scripts/ems/experiment.meta index 562cf8ec2..526683363 100644 --- a/scripts/ems/experiment.meta +++ b/scripts/ems/experiment.meta @@ -729,19 +729,21 @@ extract-phrases in: corpus-mml-postfilter=OR=word-alignment scored-corpus out: extracted-phrases rerun-on-change: max-phrase-length translation-factors reordering-factors hierarchical-rule-set extract-settings training-options script use-ghkm domain-features baseline-extract lexicalized-reordering + pass-if: mmsapt only-existence-matters: domain-features default-name: model/extract build-reordering in: extracted-phrases out: reordering-table ignore-unless: lexicalized-reordering + pass-if: mmsapt rerun-on-change: lexicalized-reordering reordering-factors default-name: model/reordering-table final-model: yes build-ttable in: extracted-phrases lexical-translation-table corpus-mml-prefilter=OR=corpus-mml-postfilter=OR=domains out: phrase-translation-table - rerun-on-change: translation-factors hierarchical-rule-set score-settings training-options script EVALUATION:report-precision-by-coverage include-word-alignment-in-rules domain-features + rerun-on-change: translation-factors hierarchical-rule-set score-settings training-options script include-word-alignment-in-rules domain-features default-name: model/phrase-table ignore-if: suffix-array mmsapt final-model: yes diff --git a/scripts/ems/experiment.perl b/scripts/ems/experiment.perl index d2dd281e8..fe8a3c9ab 100755 --- a/scripts/ems/experiment.perl +++ b/scripts/ems/experiment.perl @@ -2571,7 +2571,7 @@ sub get_config_tables { $cmd .= ":$numFF" if defined($numFF); $cmd .= " "; - $cmd .= &get_table_name_settings("reordering-factors","reordering-table",$reordering_table) if $reordering_table; + $cmd .= &get_table_name_settings("reordering-factors","reordering-table",$reordering_table) if $reordering_table && !defined($mmsapt); $cmd .= &get_table_name_settings("generation-factors","generation-table",$generation_table) if $generation_table; $cmd .= "-config $config "; diff --git a/scripts/generic/multi_moses.py b/scripts/generic/multi_moses.py index 01a38d8e6..f47cba1f5 100755 --- a/scripts/generic/multi_moses.py +++ b/scripts/generic/multi_moses.py @@ -90,9 +90,10 @@ def run_instance(cmd_base, threads, tasks, n_best=False): cmd.append('--threads') cmd.append(str(threads)) try: - # Queue of tasks instance is currently working on, limited to the number of - # threads. The queue should be kept full for optimal CPU usage. - work = Queue.Queue(maxsize=threads) + # Queue of tasks instance is currently working on, limited to the number + # of threads * 2 (minimal buffering). The queue should be kept full for + # optimal CPU usage. + work = Queue.Queue(maxsize=(threads * 2)) # Multi-threaded instance moses = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE) diff --git a/scripts/generic/score_parallel.py b/scripts/generic/score_parallel.py new file mode 100755 index 000000000..b4e4815e9 --- /dev/null +++ b/scripts/generic/score_parallel.py @@ -0,0 +1,775 @@ +#! /usr/bin/env python +# +# This file is part of moses. Its use is licensed under the GNU Lesser General +# Public License version 2.1 or, at your option, any later version. +# +# Script contributed by Precision Translation Tools. + +"""Run Moses `score` jobs in parallel. + +This script is a replacement for `score-parallel.perl`. The two are similar, +but there are differences in usage. In addition, this script can be called +directly from Python code without the need to run it as a separate process. +""" + +from __future__ import ( + absolute_import, + print_function, + unicode_literals, + ) + +__metaclass__ = type + +from argparse import ArgumentParser +from contextlib import contextmanager +from datetime import datetime +import errno +import gzip +from multiprocessing import Pool +import os +import os.path +import pipes +from shutil import rmtree +from subprocess import check_call +import sys +import tempfile + + +def get_unicode_type(): + """Return the Unicode string type appropriate to this Python version.""" + if sys.version_info.major <= 2: + # Unicode string type. In Python 2 this is the "unicode" type, + # while "str" is a binary string type. + return unicode + else: + # Unicode string type. In Python 3 this is the default "str" type. + # The binary string type is now called "bytes". + return str + + +UNICODE_TYPE = get_unicode_type() + + +class CommandLineError(Exception): + """Invalid command line.""" + + +class ProgramFailure(Exception): + """Failure, not a bug, which is reported neatly to the user.""" + + +def parse_args(): + """Parse command line arguments, return as `Namespace`.""" + parser = ArgumentParser(description=__doc__) + parser.add_argument( + '--extract-file', '-e', metavar='PATH', required=True, + help=( + "Path to input file: extract file (e.g. 'extract.sorted.gz' or " + "'extract.inv.sorted.gz'). Required.")) + parser.add_argument( + '--lex-file', '-l', metavar='PATH', required=True, + help=( + "Path to input file: lex file (e.g. 'lex.f2e' or 'lex.e2f'). " + "Required.")) + parser.add_argument( + '--output', '-o', metavar='PATH', required=True, + help=( + "Write phrase table to file PATH (e.g. 'phrase-table.half.f2e' " + "or 'phrase-table.half.e2f'). Required.")) + parser.add_argument( + '--inverse', '-i', action='store_true', + help="Inverse scoring. Defaults to direct scoring.") + parser.add_argument( + '--labels-file', '-L', metavar='PATH', + help="Also write source labels to file PATH.") + parser.add_argument( + '--parts-of-speech', '-p', metavar='PATH', + help="Also write parts-of-speech file to PATH.") + parser.add_argument( + '--flexibility-score', '-F', metavar='PATH', + help="Path to the 'flexibility_score.py' script. Defaults to none.") + parser.add_argument( + '--hierarchical', '-H', action='store_true', + help="Process hierarchical rules.") + parser.add_argument( + '--args', '-a', metavar='ARGUMENTS', + help="Additional arguments for `score` and `flexibility_score`.") + parser.add_argument( + '--sort', '-s', action='store_true', + help="Sort output file.") + parser.add_argument( + '--jobs', '-j', metavar='N', type=int, default=1, + help="Run up to N jobs in parallel. Defaults to %(default)s.") + parser.add_argument( + '--score-exe', '-x', metavar='PROGRAM', + help="Name of, or path to, the 'score' executable.") + parser.add_argument( + '--sort-command', '-S', metavar='COMMAND-LINE', + help=( + "Command line for sorting text files to standard output. " + "Must support operation as a pipe, as well as input files named " + "as command-line arguments.")) + parser.add_argument( + '--gzip-command', '-z', metavar='PROGRAM', + help="Path to a gzip or pigz executable.") + parser.add_argument( + '--verbose', '-v', action='store_true', + help="Print what's going on.") + parser.add_argument( + '--debug', '-d', action='store_true', + help="Don't delete temporary directories when done.") + return parser.parse_args() + + +def normalize_path(optional_path=None): + """Return a cleaned-up version of a given filesystem path, or None. + + Converts the path to the operating system's native conventions, and + removes redundancies like `.`. + + The return value will be `None`, an absolute path, or a relative path, + same as the argument. But it will have redundant path separators, + unnecessary detours through parent directories, and use of the current + directory "." removed. + """ + if optional_path is None: + return None + else: + path = os.path.normpath(optional_path) + path = path.replace('/', os.path.sep) + path = path.replace('\\', os.path.sep) + return path + + +def quote(path): + """Quote and escape a filename for use in a shell command. + + The Windows implementation is very limited and will break on anything + more advanced than a space. + """ + if os.name == 'posix': + return pipes.quote(path) + else: + # TODO: Improve escaping for Windows. + return '"%s"' % path + + +def sanitize_args(args): + """Check `args` for sanity, clean up, and set nontrivial defaults.""" + if args.jobs < 1: + raise CommandLineError("Number of parallel jobs must be 1 or more.") + if args.sort_command is None: + args.sort_command = find_first_executable( + ['neandersort', 'gsort', 'sort']) + if args.sort_command is None: + raise CommandLineError( + "No 'sort' command is available. " + "Choose one using the --sort-command option.") + if args.gzip_command is None: + args.gzip_command = find_first_executable(['pigz', 'gzip']) + if args.gzip_command is None: + raise CommandLineError( + "No 'gzip' or 'pigz' command is available. " + "Choose one using the --gzip-command option.") + if args.score_exe is None: + # Look for "score" executable. It may be in the current project + # directory somewhere, or in the PATH. + moses_dir = os.path.dirname(os.path.dirname( + os.path.abspath(__file__))) + args.score_exe = find_first_executable( + ['score'], + [ + moses_dir, + os.path.join(moses_dir, 'phrase-extract'), + os.path.join(moses_dir, 'binaries'), + ]) + args.extract_file = normalize_path(args.extract_file) + args.lex_file = normalize_path(args.lex_file) + args.output = normalize_path(args.output) + args.labels_file = normalize_path(args.labels_file) + args.parts_of_speech = normalize_path(args.parts_of_speech) + args.flexibility_score = normalize_path(args.flexibility_score) + args.score_exe = normalize_path(args.score_exe) + + +def add_exe_suffix(program): + """Return the full filename for an executable. + + On Windows, this adds a `.exe` suffix to the name. On other + systems, it returns the original name unchanged. + """ + if os.name == 'nt': + # Windows. + return program + '.exe' + else: + # Assume POSIX or similar. + return program + + +def find_executable(exe, extra_path=None): + """Return full path to an executable of the given name, or `None`. + + If the given name is a qualified path to an executable, it will be returned + unchanged. A qualified path where no executable is found results in a + `CommandLineError`. + """ + if extra_path is None: + extra_path = [] + + if os.path.sep in exe: + # The executable name includes a path. Only one place it can be. + if not os.path.isfile(exe) or not os.access(exe, os.X_OK): + raise CommandLineError("Not an executable: '%s'." % exe) + return exe + + for path in extra_path + os.getenv('PATH').split(os.pathsep): + full_path = os.path.join(path, exe) + if os.access(full_path, os.X_OK): + return full_path + return None + + +def find_first_executable(candidates, extra_path=None): + """Find the first available of the given candidate programs. + + :raise ProgramFailure: If none of `candidates` was found. + """ + for program in candidates: + executable = find_executable(add_exe_suffix(program), extra_path) + if executable is not None: + return executable + raise ProgramFailure( + "Could not find any of these executables in path: %s." + % ', '.join(candidates)) + + +def execute_shell(command, verbose=False): + """Run `command` string through the shell. + + Inherits environment, but sets `LC_ALL` to `C` for predictable results, + especially from sort commands. + + This uses a full-featured shell, including pipes, substitution, etc. So + remember to quote/escape arguments where appropriate! + """ + assert isinstance(command, UNICODE_TYPE), ( + "Wrong argument for execute_shell.") + if verbose: + print("Executing: %s" % command) + env = os.environ.copy() + if os.name == 'posix': + env['LC_ALL'] = 'C' + check_call(command, shell=True, env=env) + + +@contextmanager +def tempdir(keep=False): + """Context manager: temporary directory.""" + directory = tempfile.mkdtemp() + yield directory + if not keep: + rmtree(directory) + + +def make_dirs(path): + """Equivalent to `mkdir -p -- path`.""" + try: + os.makedirs(path) + except OSError as error: + if error.errno != errno.EEXIST: + raise + + +def open_file(path, mode='r'): + """Open a file, which may be gzip-compressed.""" + if path.endswith('.gz'): + return gzip.open(path, mode) + else: + return open(path, mode) + + +def count_lines(filename): + """Count the number of lines in `filename` (may be gzip-compressed).""" + count = 0 + with open_file(filename) as stream: + for _ in stream: + count += 1 + return count + + +def set_temp_dir(): + """Set temporary directory to `$MOSES_TEMP_DIR`, if set. + + Create the directory if necessary. + """ + temp_dir = os.getenv('MOSES_TEMP_DIR') + if temp_dir is not None: + make_dirs(temp_dir) + tempfile.tempdir = temp_dir + + +def strip_newline(line): + """Remove trailing carriage return and/or line feed, if present.""" + if line.endswith('\n'): + line = line[:-1] + if line.endswith('\r'): + line = line[:-1] + return line + + +def open_chunk_file(split_dir, chunk_number): + """Open a file to write one chunk of the extract file.""" + return open_file( + os.path.join(split_dir, 'extract.%d.gz' % chunk_number), 'w') + + +def name_context_chunk_file(split_dir, chunk_number): + """Compose file name for one chunk of the extract context file.""" + return os.path.join( + split_dir, 'extract.context.%d.gz' % chunk_number) + + +def extract_source_phrase(line): + """Extract the source phrase from an extract-file line.""" + return line.split(b'|||', 1)[0] + + +def cut_context_file(last_source_phrase, chunk_file, last_line, + context_stream): + """Write one chunk of extract context file into its own file. + + :param last_source_phrase: Last source phrase that should be in the + chunk. Stop processing after this source phrase. + :param chunk_file: Path to the extract context file for this chunk. + :param last_line: Previously read line that may still need writing. + :param context_stream: Extract context file, opened for reading. + :return: Last line read from `context_stream`. This line will still + need processing. + """ + # TODO: Use open_file. + with gzip.open(chunk_file, 'w') as chunk: + if last_line is not None: + chunk.write('%s\n' % last_line) + + # Are we processing our last source phrase yet? + on_last_source_phrase = False + + # Write all lines in context file until we meet last source phrase + # in extract file. + for line in context_stream: + # Reading from a gzip file returns lines *including the newline*. + # Either way, we want to ignore carriage returns as well. + line = strip_newline(line) + source_phrase = extract_source_phrase(line) + if on_last_source_phrase and source_phrase != last_source_phrase: + # First new source phrase after our last one. We're done. + return line + else: + # Still adding lines to our chunk. + chunk.write('%s\n' % line) + if source_phrase == last_source_phrase: + # We're on our last source phrase now. + on_last_source_phrase = True + + +def split_extract_files(split_dir, extract_file, extract_context_file=None, + jobs=1): + """Split extract file into chunks, so we can process them in parallel. + + :param split_dir: A temporary directory where this function can write + temporary files. The caller must ensure that this directory will be + cleaned up after it's done with the files. + :return: An iterable of tuples. Each tuple hols a partial extract file, + and the corresponding context file. The files may be in `split_dir`, + or there may just be the original extract file. + """ + if jobs == 1: + # No splitting needed. Read the original file(s). + return [(extract_file, extract_context_file)] + + # Otherwise: split files. + files = [] + num_lines = count_lines(extract_file) + chunk_size = (num_lines + jobs - 1) / jobs + assert isinstance(chunk_size, int) + + line_count = 0 + chunk_number = 0 + prev_source_phrase = None + last_line_context = None + extract_stream = open_file(extract_file) + chunk_file = open_chunk_file(split_dir, chunk_number) + if extract_context_file is None: + chunk_context_file = None + if extract_context_file is not None: + context_stream = open_file(extract_context_file) + + for line in extract_stream: + line_count += 1 + line = line.decode('utf-8') + line = strip_newline(line) + if line_count >= chunk_size: + # At or over chunk size. Cut off at next source phrase change. + source_phrase = extract_source_phrase(line) + if prev_source_phrase is None: + # Start looking for a different source phrase. + prev_source_phrase = source_phrase + elif source_phrase == prev_source_phrase: + # Can't cut yet. Still working on the same source phrase. + pass + else: + # Hit first new source phrase after chunk limit. Cut new + # file(s). + chunk_file.close() + if extract_context_file is not None: + chunk_context_file = name_context_chunk_file( + split_dir, chunk_number) + last_line_context = cut_context_file( + prev_source_phrase, chunk_context_file, + last_line_context, context_stream) + files.append((chunk_file.name, chunk_context_file)) + + # Start on new chunk. + prev_source_phrase = None + line_count = 0 + chunk_number += 1 + chunk_file = open_chunk_file(split_dir, chunk_number) + chunk_file.write(('%s\n' % line).encode('utf-8')) + + chunk_file.close() + if extract_context_file is not None: + chunk_context_file = name_context_chunk_file(split_dir, chunk_number) + last_line_context = cut_context_file( + prev_source_phrase, chunk_number, last_line_context, + context_stream) + files.append((chunk_file.name, chunk_context_file)) + return files + + +def compose_score_command(extract_file, context_file, half_file, + flex_half_file, args): + """Compose command line text to run one instance of `score`. + + :param extract_file: One chunk of extract file. + :param context_file: If doing flexibility scoring, one chunk of + extract context file. Otherwise, None. + :param half_file: ??? + :param flex_half_file: ??? + :param args: Arguments namespace. + """ + command = [ + args.score_exe, + extract_file, + args.lex_file, + half_file, + ] + if args.args not in (None, ''): + command.append(args.args) + other_args = build_score_args(args) + if other_args != '': + command.append(other_args) + if context_file is not None: + command += [ + '&&', + find_first_executable(['bzcat']), + '|', + quote(args.flexibility_score), + quote(context_file), + ] + if args.inverse: + command.append('--Inverse') + if args.hierarchical: + command.append('--Hierarchical') + command += [ + '|', + quote(args.gzip_command), + '-c', + '>%s' % quote(flex_half_file), + ] + return ' '.join(command) + + +def score_parallel(split_dir, file_pairs, args): + """Run the `score` command in parallel. + + :param split_dir: Temporary directory where we can create split files. + :param file_pairs: Sequence of tuples for the input files, one tuple + per chunk of the work. Each tuple consists of a partial extract + file, and optionally a partial extract context file. + :param args: Arguments namespace. + :return: A list of tuples. Each tuple contains two file paths. The first + is for a partial half-phrase-table file. The second is for the + corresponding partial flex file, if a context file is given; or + `None` otherwise. + """ + partial_files = [] + # Pool of worker processes for executing the partial "score" invocations + # concurrently. + pool = Pool(args.jobs) + try: + for chunk_num, file_pair in enumerate(file_pairs): + half_file = os.path.join( + split_dir, 'phrase-table.half.%06d.gz' % chunk_num) + extract_file, context_file = file_pair + if context_file is None: + flex_half_file = None + else: + flex_half_file = os.path.join( + split_dir, 'phrase-table.half.%06d.flex.gz' % chunk_num) + # Pickling of arguments for the pool is awkward on Windows, so + # keep them simple. Compose the command line in the parent + # process, then hand them to worker processes which execute them. + command_line = compose_score_command( + extract_file, context_file, half_file, flex_half_file, args) + pool.apply_async( + execute_shell, (command_line, ), {'verbose': args.verbose}) + partial_files.append((half_file, flex_half_file)) + pool.close() + except BaseException: + pool.terminate() + raise + finally: + pool.join() + return partial_files + + +def merge_and_sort(files, output, sort_command=None, gzip_exe=None, + verbose=False): + """Merge partial files. + + :param files: List of partial half-phrase-table files. + :param output: Path for resulting combined phrase-table file. + """ +# TODO: The Perl code mentioned "sort" and "flexibility_score" here. +# What do we do with those? + + # Sort whether we're asked to or not, as a way of combining the input + # files. + if sort_command == 'neandersort': + # Neandersort transparently decompresses input and compresses output. + check_call([ + 'neandersort', + '-o', output, + ] + files) + else: + command = ( + "%(gzip)s -c -d %(files)s | " + "%(sort)s | " + "%(gzip)s -c >>%(output)s" + % { + 'gzip': quote(gzip_exe), + 'sort': sort_command, + 'files': ' '.join(map(quote, files)), + 'output': quote(output), + }) + execute_shell(command, verbose=verbose) + + +def build_score_args(args): + """Compose command line for the `score` program.""" + command_line = [] + if args.labels_file: + command_line += [ + '--SourceLabels', + '--SourceLabelCountsLHS', + '--SourceLabelSet', + ] + if args.parts_of_speech: + command_line.append('--PartsOfSpeech') + if args.inverse: + command_line.append('--Inverse') + if args.args is not None: + command_line.append(args.args) + return ' '.join(command_line) + + +def list_existing(paths): + """Return, in the same order, those of the given files which exist.""" + return filter(os.path.exists, paths) + + +def compose_coc_path_for(path): + """Compose COC-file path for the given file.""" + return '%s.coc' % path + + +def read_cocs(path): + """Read COC file at `path`, return contents as tuple of ints.""" + with open(path) as lines: + return tuple( + int(line.rstrip('\r\n')) + for line in lines + ) + + +def add_cocs(original, additional): + """Add two tuples of COCs. Extend as needed.""" + assert not (original is None and additional is None), "No COCs to add!" + if original is None: + return additional + elif additional is None: + return original + else: + common = tuple(lhs + rhs for lhs, rhs in zip(original, additional)) + return ( + common + + tuple(original[len(common):]) + + tuple(additional[len(common):])) + + +def merge_coc(files, output): + """Merge COC files for the given partial files. + + Each COC file is a series of integers, one per line. This reads them, and + adds them up line-wise into one file of the same format: the sum of the + numbers the respective files have at line 1, the sum of the numbers the + respective files have at line 2, and so on. + """ + assert len(files) > 0, "No partial files - no work to do." + extract_files = [extract_file for extract_file, _ in files] + if not os.path.exists(compose_coc_path_for(extract_files[0])): + # Nothing to merge. + return + totals = None +# TODO: Shouldn't we just fail if any of these files is missing? + for coc_path in list_existing(map(compose_coc_path_for, extract_files)): + totals = add_cocs(totals, read_cocs(coc_path)) + + # Write to output file. + with open(output, 'w') as output_stream: + for entry in totals: + output_stream.write('%d\n' % entry) + + +def suffix_line_numbers(infile, outfile): + """Rewrite `infile` to `outfile`; suffix line number to each line. + + The line number is zero-based, and separated from the rest of the line + by a single space. + """ + temp_file = '%s.numbering' % outfile + with open(infile, 'r') as instream, open(outfile, 'w') as outstream: + line_no = 0 + for line in instream: + outstream.write(line) + outstream.write(' %d\n' % line_no) + line_no += 1 + os.rename(temp_file, outfile) + + +def compose_source_labels_path_for(path): + """Return source labels file path for given file.""" + return '%s.syntaxLabels.src' % path + + +def merge_numbered_files(inputs, output, header_lines, sort_command, + verbose=False): + """Sort and merge files `inputs`, add header and line numbers. + + :param inputs: Iterable of input files. + :param output: Output file. + :header_lines: Iterable of header lines. + :sort_command: Command line for sorting input files. + """ + sort_temp = '%s.sorting' % output + with open(sort_temp, 'w') as stream: + for line in header_lines: + stream.write(line) + stream.write('\n') + execute_shell( + "%s %s >>%s" % ( + sort_command, + ' '.join(map(quote, inputs)), + quote(sort_temp)), + verbose=verbose) + suffix_line_numbers(sort_temp, output) + + +def merge_source_labels(files, output, sort_command, verbose=False): + """Merge source labels files.""" +# TODO: Shouldn't we just fail if any of these files is missing? + labels_files = list_existing(map(compose_source_labels_path_for, files)) + header = [ + 'GlueTop', + 'GlueX', + 'SSTART', + 'SEND', + ] + merge_numbered_files( + labels_files, output, header, sort_command, verbose=verbose) + + +def compose_parts_of_speech_path_for(path): + """Return parts-of-speech file path for given file.""" + return '%s.partsOfSpeech' % path + + +def merge_parts_of_speech(files, output, sort_command, verbose=False): + """Merge parts-of-speech files into output.""" +# TODO: Shouldn't we just fail if any of these files is missing? + parts_files = list_existing(map(compose_parts_of_speech_path_for, files)) + header = [ + 'SSTART', + 'SEND', + ] + merge_numbered_files( + parts_files, output, header, sort_command, verbose=verbose) + + +def main(): + """Command-line entry point. Marshals and forwards to `score_parallel`.""" + args = parse_args() + sanitize_args(args) + set_temp_dir() + + if args.flexibility_score is None: + extract_context_file = None + else: + extract_context_file = args.extract_file.replace( + 'extract.', 'extract.context.') + + if args.verbose: + print("Started %s." % datetime.now()) + print("Using '%s' for gzip." % args.gzip_command) + + with tempdir(args.debug) as split_dir: + extract_files = split_extract_files( + split_dir, args.extract_file, + extract_context_file=extract_context_file, jobs=args.jobs) + + scored_files = score_parallel(split_dir, extract_files, args) + + if args.verbose: + sys.stderr.write("Finished score %s.\n" % datetime.now()) + +# TODO: Pass on "sort" and "flexibility-score" arguments? + merge_and_sort( + [phrase_chunk for phrase_chunk, _ in scored_files], args.output, + sort_command=args.sort_command, gzip_exe=args.gzip_command, + verbose=args.verbose) + merge_coc(extract_files, compose_coc_path_for(args.output)) + + if not args.inverse and args.labels_file is not None: + if args.verbose: + print("Merging source labels files.") + merge_source_labels( + extract_files, args.labels_file, + sort_command=args.sort_command, verbose=args.verbose) + + if not args.inverse and args.parts_of_speech is not None: + if args.verbose: + print("Merging parts-of-speech files.") + merge_parts_of_speech( + extract_files, args.parts_of_speech, + sort_command=args.sort_command, verbose=args.verbose) + + +if __name__ == '__main__': + try: + main() + except ProgramFailure as error: + sys.stderr.write('%s\n' % error) + sys.exit(1) + except CommandLineError as error: + sys.stderr.write("Command line error: %s\n" % error) + sys.exit(2) diff --git a/scripts/training/train-model.perl b/scripts/training/train-model.perl index c69bceeaa..76f661712 100755 --- a/scripts/training/train-model.perl +++ b/scripts/training/train-model.perl @@ -2141,7 +2141,8 @@ sub create_ini { # sum up... $feature_spec .= "$phrase_table_impl_name name=TranslationModel$i num-features=$basic_weight_count path=$file input-factor=$input_factor output-factor=$output_factor"; - $feature_spec .= " L1=$___F L2=$___E " if defined($_MMSAPT); # extra settings for memory mapped suffix array phrase table + $feature_spec .= " L1=$___F L2=$___E" if defined($_MMSAPT); # extra settings for memory mapped suffix array phrase table + $feature_spec .= " lr-func=LexicalReordering0" if defined($_MMSAPT) && $i==0 && $REORDERING_LEXICAL; $feature_spec .= "\n"; unless ($phrase_table_impl==11) { # suffix array provides its weights at first iteration $weight_spec .= "TranslationModel$i="; @@ -2222,7 +2223,7 @@ sub create_ini { $table_file .= "."; $table_file .= $model->{"filename"}; $table_file .= ".gz"; - $feature_spec .= "LexicalReordering name=LexicalReordering$i num-features=".$model->{"numfeatures"}." type=".$model->{"config"}." input-factor=$input_factor output-factor=$output_factor path=$table_file".(defined($_LEXICAL_REORDERING_DEFAULT_SCORES)?" default-scores=$_LEXICAL_REORDERING_DEFAULT_SCORES":"")."\n"; + $feature_spec .= "LexicalReordering name=LexicalReordering$i num-features=".$model->{"numfeatures"}." type=".$model->{"config"}." input-factor=$input_factor output-factor=$output_factor".((defined($_MMSAPT)&&$i==0)?"":" path=$table_file").(defined($_LEXICAL_REORDERING_DEFAULT_SCORES)?" default-scores=$_LEXICAL_REORDERING_DEFAULT_SCORES":"")."\n"; $weight_spec .= "LexicalReordering$i="; for(my $j=0;$j<$model->{"numfeatures"};$j++) { $weight_spec .= " 0.3"; } $weight_spec .= "\n"; |