Added flags to mgiza to explicitly specify input *.classes files.

The new flags are -sourcevocabularyclasses and -targetvocabularyclasses

5 files changed, 42 insertions, 18 deletions

diff --git a/mgizapp/src/Globals.h b/mgizapp/src/Globals.h
index 3fa17d9..693a117 100644
--- a/mgizapp/src/Globals.h
+++ b/mgizapp/src/Globals.h
@@ -32,6 +32,7 @@ extern float PROB_SMOOTH,MINCOUNTINCREASE;
 extern bool Verbose, Log, Peg, Transfer, Transfer2to3, useDict ;
 extern string Prefix, LogFilename, OPath, 
   SourceVocabFilename, TargetVocabFilename, CorpusFilename, TestCorpusFilename, 
+  SourceVocabClassesFilename, TargetVocabClassesFilename,
   t_Filename, a_Filename, p0_Filename, d_Filename, n_Filename, dictionary_Filename;
 extern ofstream logmsg ;
 extern Mutex logmsg_lock;
diff --git a/mgizapp/src/d4norm.cxx b/mgizapp/src/d4norm.cxx
index 7a1da31..a790a62 100644
--- a/mgizapp/src/d4norm.cxx
+++ b/mgizapp/src/d4norm.cxx
@@ -56,6 +56,7 @@ GLOBAL_PARAMETER(WordIndex, MAX_FERTILITY, "MAX_FERTILITY",
 using namespace std;
 string Prefix, LogFilename, OPath, Usage, SourceVocabFilename,
 		TargetVocabFilename, CorpusFilename, TestCorpusFilename, t_Filename,
+		SourceVocabClassesFilename, TargetVocabClassesFilename,
 		a_Filename, p0_Filename, d_Filename, n_Filename, dictionary_Filename;
 
 
@@ -74,11 +75,11 @@ int main(int argc, char* argv[]){
 	fTrainVcbList.setName(argv[2]);
 	eTrainVcbList.readVocabList();
 	fTrainVcbList.readVocabList();
-	string evcbcls = argv[1];
-	string fvcbcls = argv[2];
-	evcbcls += ".classes";
-	fvcbcls += ".classes";
-	d4m.makeWordClasses(eTrainVcbList, fTrainVcbList, evcbcls.c_str(), fvcbcls.c_str(),eTrainVcbList,fTrainVcbList);
+	SourceVocabClassesFilename = argv[1];
+	TargetVocabClassesFilename = argv[2];
+	SourceVocabClassesFilename += ".classes";
+	TargetVocabClassesFilename += ".classes";
+	d4m.makeWordClasses(eTrainVcbList, fTrainVcbList, SourceVocabClassesFilename.c_str(), TargetVocabClassesFilename.c_str(),eTrainVcbList,fTrainVcbList);
 	// Start iteration:
 	for(int i =4; i< argc ; i++){
 		string name = argv[i];
diff --git a/mgizapp/src/hmmnorm.cxx b/mgizapp/src/hmmnorm.cxx
index 9d737c5..2643102 100644
--- a/mgizapp/src/hmmnorm.cxx
+++ b/mgizapp/src/hmmnorm.cxx
@@ -55,6 +55,7 @@ GLOBAL_PARAMETER(WordIndex, MAX_FERTILITY, "MAX_FERTILITY",
 using namespace std;
 string Prefix, LogFilename, OPath, Usage, SourceVocabFilename,
 		TargetVocabFilename, CorpusFilename, TestCorpusFilename, t_Filename,
+		SourceVocabClassesFilename, TargetVocabClassesFilename,
 		a_Filename, p0_Filename, d_Filename, n_Filename, dictionary_Filename;
 
 
@@ -84,11 +85,11 @@ int main(int argc, char* argv[]){
 	model2 m2(m1, aTable, aCountTable);
 	WordClasses french,english;
 	hmm h(m2,english,french);
-	string evcbcls = argv[1];
-	string fvcbcls = argv[2];
-	evcbcls += ".classes";
-	fvcbcls += ".classes";
-	h.makeWordClasses(m1.Elist, m1.Flist, evcbcls.c_str(), fvcbcls.c_str());
+	SourceVocabClassesFilename = argv[1];
+	TargetVocabClassesFilename = argv[2];
+	SourceVocabClassesFilename += ".classes";
+	TargetVocabClassesFilename += ".classes";
+	h.makeWordClasses(m1.Elist, m1.Flist, SourceVocabClassesFilename.c_str(), TargetVocabClassesFilename.c_str());
 	string base = argv[4];
 	string baseA = base+".alpha";
 	string baseB = base+".beta";
diff --git a/mgizapp/src/main.cpp b/mgizapp/src/main.cpp
index cd44526..d3a23c9 100644
--- a/mgizapp/src/main.cpp
+++ b/mgizapp/src/main.cpp
@@ -98,6 +98,7 @@ Vector<map< pair<int,int>,char > > ReferenceAlignment;
 bool useDict = false;
 string CoocurrenceFile;
 string Prefix, LogFilename, OPath, Usage, SourceVocabFilename,
+		SourceVocabClassesFilename(""), TargetVocabClassesFilename(""),
 		TargetVocabFilename, CorpusFilename, TestCorpusFilename, t_Filename,
 		a_Filename, p0_Filename, d_Filename, n_Filename, dictionary_Filename;
 
@@ -250,8 +251,8 @@ void printDecoderConfigFile() {
 	decoder << "Target.vcb = " << TargetVocabFilename << '\n';
 	//  decoder << "Source.classes = " << SourceVocabFilename + ".classes" << '\n';
 	//  decoder << "Target.classes = " << TargetVocabFilename + ".classes" <<'\n';
-	decoder << "Source.classes = " << SourceVocabFilename+".classes" << '\n';
-	decoder << "Target.classes = " << TargetVocabFilename + ".classes" <<'\n';
+	decoder << "Source.classes = " << SourceVocabClassesFilename << '\n';
+	decoder << "Target.classes = " << TargetVocabClassesFilename <<'\n';
 	p=Prefix + ".fe0_"+ /*lastModelName*/"3" + ".final";
 	decoder << "FZeroWords       = " <<stripPath(p.c_str()) << '\n';
 
@@ -836,8 +837,8 @@ double StartTraining(int&result) {
 			if (HMM_Iterations > 0 && (restart < 2 || restart == 4 || restart == 5 || restart == 6)) {
 				cout << "NOTE: I am doing iterations with the HMM model!\n";
 				
-				h.makeWordClasses(m1.Elist, m1.Flist, SourceVocabFilename
-						+".classes", TargetVocabFilename+".classes");
+				h.makeWordClasses(m1.Elist, m1.Flist, SourceVocabClassesFilename
+						, TargetVocabClassesFilename);
 				if(restart != 6) h.initialize_table_uniformly(*corpus);
 				
 				if(Model3_Iterations == 0 && Model4_Iterations == 0 &&
@@ -873,8 +874,8 @@ double StartTraining(int&result) {
 				errors=m3.errorsAL();
 			}
 			if(restart >= 7 && hmmvalid){
-				h.makeWordClasses(m1.Elist, m1.Flist, SourceVocabFilename
-						+".classes", TargetVocabFilename+".classes");
+				h.makeWordClasses(m1.Elist, m1.Flist, SourceVocabClassesFilename
+						, TargetVocabClassesFilename);
 			}
 			if (HMM_Iterations>0 || restart == 7)
 				m3.setHMM(&h);
@@ -960,6 +961,18 @@ int main(int argc, char* argv[]) {
 			"target vocabulary file name",
 			TargetVocabFilename,-1));
 	getGlobalParSet().insert(new Parameter<string>(
+			"Source Vocabulary Classes",
+			ParameterChangedFlag,
+			"source vocabulary classes file name",
+			SourceVocabClassesFilename,
+			PARLEV_INPUT));
+	getGlobalParSet().insert(new Parameter<string>(
+			"Target Vocabulary Classes",
+			ParameterChangedFlag,
+			"target vocabulary classes file name",
+			TargetVocabClassesFilename,
+			PARLEV_INPUT));
+	getGlobalParSet().insert(new Parameter<string>(
 			"C",
 			ParameterChangedFlag,
 			"training corpus file name",
@@ -1092,6 +1105,14 @@ int main(int argc, char* argv[]) {
 	// 
 	parseArguments(argc, argv);
 
+	if (SourceVocabClassesFilename=="") {
+	  makeSetCommand("sourcevocabularyclasses",SourceVocabFilename+".classes",getGlobalParSet(),2);
+	}
+
+	if (TargetVocabClassesFilename=="") {
+	  makeSetCommand("targetvocabularyclasses",TargetVocabFilename+".classes",getGlobalParSet(),2);
+	}	
+
     // Determine number of threads
 
     if(NCPUS == 0){
diff --git a/mgizapp/src/model3.cpp b/mgizapp/src/model3.cpp
index 797dd58..ec3c701 100644
--- a/mgizapp/src/model3.cpp
+++ b/mgizapp/src/model3.cpp
@@ -366,8 +366,8 @@ int model3::viterbi(int noIterationsModel3, int noIterationsModel4,
 		d4m.readProbTable(previous_d4model.c_str(),previous_d4model_1.c_str());
 	}
 	if(h==NULL)
-	d4m.makeWordClasses(Elist, Flist, SourceVocabFilename+".classes",
-			TargetVocabFilename+".classes",Elist,Flist);
+	d4m.makeWordClasses(Elist, Flist, SourceVocabClassesFilename,
+			TargetVocabClassesFilename,Elist,Flist);
 	
 	d5model d5m(d4m);
 	//d5m.makeWordClasses(Elist, Flist, SourceVocabFilename+".classes",