Add scorer factory. Fix compile error in Optimizer

git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1706 1f5c12ca-751b-0410-a591-d2e778427230

11 files changed, 138 insertions, 174 deletions

diff --git a/mert/BleuScorer.cpp b/mert/BleuScorer.cpp
index f3cbdd320..d07f79008 100644
--- a/mert/BleuScorer.cpp
+++ b/mert/BleuScorer.cpp
@@ -1,4 +1,4 @@
-#include "BleuScorer.h"
+#include "Scorer.h"
 
 const int BleuScorer::LENGTH = 4;
 
diff --git a/mert/BleuScorer.h b/mert/BleuScorer.h
deleted file mode 100644
index ce541345a..000000000
--- a/mert/BleuScorer.h
+++ /dev/null
@@ -1,102 +0,0 @@
-#ifndef __BLEUSCORER_H__
-#define __BLEUSCORER_H__
-
-#include <cctype>
-#include <cmath>
-#include <ctime>
-#include <stdexcept>
-#include <fstream>
-#include <iostream>
-#include <iterator>
-#include <map>
-#include <sstream>
-#include <string>
-#include <vector>
-
-#include "Util.h"
-#include "Scorer.h"
-#include "ScoreData.h"
-
-using namespace std;
-
-enum BleuReferenceLengthStrategy { AVERAGE, SHORTEST, CLOSEST };
-
-
-/**
-  * Bleu scoring
- **/
-class BleuScorer: public StatisticsBasedScorer {
-	public:
-		BleuScorer() : StatisticsBasedScorer("BLEU"),_refLengthStrategy(SHORTEST) {}
-		virtual void setReferenceFiles(const vector<string>& referenceFiles);
-		virtual void prepareStats(int sid, const string& text, ScoreStats& entry);
-		static const int LENGTH;	
-    
-    protected:
-        float calculateScore(const vector<int>& comps);
-		
-	private:
-		//no copy
-		BleuScorer(const BleuScorer&);
-		BleuScorer& operator=(const BleuScorer&);
-				
-
-		//Used to construct the ngram map
-		struct CompareNgrams {
-			int operator() (const vector<int>& a, const vector<int>& b) {
-				/*
-				cerr << "compare:";
-				copy(a.begin(), a.end(), ostream_iterator<int>(cerr," "));
-				cerr << " with ";
-				copy(b.begin(), b.end(), ostream_iterator<int>(cerr," "));
-				cerr << endl;
-				*/
-				size_t i;
-				size_t as = a.size();
-				size_t bs = b.size();
-				for (i = 0; i < as && i < bs; ++i) {
-					if (a[i] < b[i]) {
-						//cerr << "true" << endl;
-						return true;
-					}
-					if (a[i] > b[i]) {
-						//cerr << "false" << endl;
-						return false;
-					}
-				}
-				//entries are equal, shortest wins
-				/*if (as < bs) {
-						cerr << "true" << endl;
-				} else {
-						cerr << "false" << endl;
-				}*/
-				return as < bs;;
-			}
-		};
-
-		typedef map<vector<int>,int,CompareNgrams> counts_t;
-		typedef map<vector<int>,int,CompareNgrams>::iterator counts_it;
-
-		typedef vector<counts_t*> refcounts_t;
-
-		size_t countNgrams(const string& line, counts_t& counts, unsigned int n);
-
-		void dump_counts(counts_t& counts) {
-			for (counts_it i = counts.begin(); i != counts.end(); ++i) {
-				cerr << "(";
-				copy(i->first.begin(), i->first.end(), ostream_iterator<int>(cerr," "));
-				cerr << ") " << i->second << ", ";
-			}
-			cerr << endl;
-		} 
-		BleuReferenceLengthStrategy _refLengthStrategy;
-		
-		// data extracted from reference files
-		refcounts_t _refcounts;
-		vector<vector<size_t> > _reflengths;
-};
-
-
-
-
-#endif //__BLEUSCORER_H
diff --git a/mert/Makefile b/mert/Makefile
index eb52da940..5c2d5cdf5 100755
--- a/mert/Makefile
+++ b/mert/Makefile
@@ -27,7 +27,7 @@ test_scorer
 clean: 
 	rm -f *.o
 
-%.o : %.cpp %.h
+%.o : %.cpp
 	$(GCC) -c $(CFLAGS) $< -o $@
 
 feature_extractor: $(OBJS) feature_extractor.cpp
diff --git a/mert/Optimizer.cpp b/mert/Optimizer.cpp
index 43973e27b..c4520272b 100644
--- a/mert/Optimizer.cpp
+++ b/mert/Optimizer.cpp
@@ -1,14 +1,16 @@
 #include <cassert>
-#include "Optimizer.h"
 #include <vector>
-#include<list>
+#include <limits>
+#include <list>
 #include <cfloat>
 #include <iostream>
 
+#include "Optimizer.h"
+
 using namespace std;
 
-static const float MINFLOAT=numeric_limits<float>::min();
-static const float MAXFLOAT=numeric_limits<float>::max();
+static const float MIN_FLOAT=numeric_limits<float>::min();
+static const float MAX_FLOAT=numeric_limits<float>::max();
 
 enum OptType{POWELL=0,NOPTIMIZER};//Add new optimizetr here
 
@@ -74,7 +76,7 @@ statscore Optimizer::GetStatScore(const Point& param)const{
 /**compute the intersection of 2 lines*/
 float intersect (float m1, float b1,float m2,float b2){
   if(m1==m2)
-    return MAXFLOAT;//parrallel lines
+    return MAX_FLOAT;//parrallel lines
   return((b2-b1)/(m1-m2));
 }
 
@@ -85,7 +87,7 @@ statscore Optimizer::LineOptimize(const Point& origin,const Point& direction,Poi
   typedef pair<float,vector<unsigned> > threshold;  
   list<threshold> thresholdlist;
   
-  thresholdlist.push_back(pair<float,vector<unsigned> >(MINFLOAT,vector<unsigned>()));
+  thresholdlist.push_back(pair<float,vector<unsigned> >(MIN_FLOAT,vector<unsigned>()));
 
   for(int S=0;S<size();S++){
     //first we determine the translation with the best feature score for each sentence and each value of x
@@ -111,7 +113,7 @@ statscore Optimizer::LineOptimize(const Point& origin,const Point& direction,Poi
 	index=it->second;//the highest line is the one with he highest f0
       }
     --it;//we went one step too far in the while loop
-    onebest.push_back(pair<float,unsigned>(MINFLOAT,index));//first 1best is the lowest gradient. 
+    onebest.push_back(pair<float,unsigned>(MIN_FLOAT,index));//first 1best is the lowest gradient. 
     //now we look for the intersections points indicating a change of 1 best
     //we use the fact that the function is convex, which means that the gradient can only go up   
     while(it!=gradient.end()){
@@ -199,7 +201,7 @@ statscore Optimizer::LineOptimize(const Point& origin,const Point& direction,Poi
   //last thing to do is compute the Stat score (ie BLEU) and find the minimum
   list<threshold>::iterator best;
   list<threshold>::iterator lit2;
-  statscore bestscore=MINFLOAT;
+  statscore bestscore=MIN_FLOAT;
   for(lit2=thresholdlist.begin();lit2!=thresholdlist.end();lit2){
     assert(lit2->second.size()==FData->size());
     statscore cur=GetStatScore(lit2->second);
@@ -222,7 +224,7 @@ void  Optimizer::Get1bests(const Point& P,vector<unsigned>& bests)const{
   bests.resize(size());
   
   for(unsigned i=0;i<size();i++){
-    float bestfs=MINFLOAT;
+    float bestfs=MIN_FLOAT;
     unsigned idx=0;
     unsigned j;
     for(j=0;j<FData->get(i).size();j++){
@@ -252,8 +254,8 @@ statscore Optimizer::Run(Point& P)const{
 }
 statscore SimpleOptimizer::TrueRun(Point& P)const{
  
-  statscore prevscore=MAXFLOAT;
-  statscore bestscore=MAXFLOAT;
+  statscore prevscore=MAX_FLOAT;
+  statscore bestscore=MAX_FLOAT;
   do{
     Point  best;
     Point  linebest;
diff --git a/mert/Optimizer.h b/mert/Optimizer.h
index d479c5020..c3000e52c 100644
--- a/mert/Optimizer.h
+++ b/mert/Optimizer.h
@@ -1,6 +1,6 @@
 #ifndef OPTIMIZER_H
 #define OPTIMIZER_H
-#include<vector>
+#include <vector>
 #include "FeatureStats.h"
 #include "FeatureData.h"
 #include "FeatureArray.h"
diff --git a/mert/PerScorer.cpp b/mert/PerScorer.cpp
index be5e29aef..d766874a9 100644
--- a/mert/PerScorer.cpp
+++ b/mert/PerScorer.cpp
@@ -1,4 +1,4 @@
-#include "PerScorer.h"
+#include "Scorer.h"
 
 
 void PerScorer::setReferenceFiles(const vector<string>& referenceFiles) {
diff --git a/mert/PerScorer.h b/mert/PerScorer.h
deleted file mode 100644
index 868326638..000000000
--- a/mert/PerScorer.h
+++ /dev/null
@@ -1,50 +0,0 @@
-#ifndef __PERSCORER_H__
-#define __PERSCORER_H__
-
-#include <stdexcept>
-#include <fstream>
-#include <iostream>
-#include <iterator>
-#include <map>
-#include <set>
-#include <sstream>
-#include <string>
-#include <vector>
-
-#include "Util.h"
-#include "Scorer.h"
-#include "ScoreData.h"
-
-using namespace std;
-
-
-/**
-  * Implementation of position-independent word error rate. This is defined
-  * as 1 - (correct - max(0,output_length - ref_length)) / ref_length
-  * In fact, we ignore the " 1 - " so that it can be maximised.
- **/
-class PerScorer: public StatisticsBasedScorer {
-	public:
-		PerScorer() : StatisticsBasedScorer("PER") {}
-		virtual void setReferenceFiles(const vector<string>& referenceFiles);
-		virtual void prepareStats(int sid, const string& text, ScoreStats& entry);
-
-    protected:
-        
-        virtual float calculateScore(const vector<int>& comps) ;
-		
-	private:
-        
-		//no copy
-		PerScorer(const PerScorer&);
-		PerScorer& operator=(const PerScorer&);
-				
-		// data extracted from reference files
-		vector<size_t> _reflengths;
-        vector<multiset<int> > _reftokens;
-};
-
-
-
-
-#endif //__PERSCORER_H
diff --git a/mert/Scorer.cpp b/mert/Scorer.cpp
index 329a401d8..400c857ff 100644
--- a/mert/Scorer.cpp
+++ b/mert/Scorer.cpp
@@ -1,7 +1,5 @@
 #include "Scorer.h"
 
-
-
 void  StatisticsBasedScorer::score(const candidates_t& candidates, const diffs_t& diffs,
             scores_t& scores) {
 	if (!_scoreData) {
diff --git a/mert/Scorer.h b/mert/Scorer.h
index cde8a238f..294cb8533 100644
--- a/mert/Scorer.h
+++ b/mert/Scorer.h
@@ -2,7 +2,11 @@
 #define __SCORER_H__
 
 #include <algorithm>
+#include <cmath>
 #include <iostream>
+#include <iterator>
+#include <set>
+#include <sstream>
 #include <stdexcept>
 #include <string>
 #include <vector>
@@ -29,7 +33,6 @@ class Scorer {
 		
 		Scorer(const string& name): _name(name), _scoreData(0),_preserveCase(false)  {}
 
-
 		/**
 		  * set the reference files. This must be called before prepareStats.
 		  **/
@@ -152,4 +155,120 @@ class StatisticsBasedScorer : public Scorer {
 };
 
 
+enum BleuReferenceLengthStrategy { AVERAGE, SHORTEST, CLOSEST };
+
+
+/**
+  * Bleu scoring
+ **/
+class BleuScorer: public StatisticsBasedScorer {
+	public:
+		BleuScorer() : StatisticsBasedScorer("BLEU"),_refLengthStrategy(SHORTEST) {}
+		virtual void setReferenceFiles(const vector<string>& referenceFiles);
+		virtual void prepareStats(int sid, const string& text, ScoreStats& entry);
+		static const int LENGTH;	
+    
+    protected:
+        float calculateScore(const vector<int>& comps);
+		
+	private:
+		//no copy
+		BleuScorer(const BleuScorer&);
+		BleuScorer& operator=(const BleuScorer&);
+				
+
+		//Used to construct the ngram map
+		struct CompareNgrams {
+			int operator() (const vector<int>& a, const vector<int>& b) {
+				size_t i;
+				size_t as = a.size();
+				size_t bs = b.size();
+				for (i = 0; i < as && i < bs; ++i) {
+					if (a[i] < b[i]) {
+						//cerr << "true" << endl;
+						return true;
+					}
+					if (a[i] > b[i]) {
+						//cerr << "false" << endl;
+						return false;
+					}
+				}
+				//entries are equal, shortest wins
+				return as < bs;;
+			}
+		};
+
+		typedef map<vector<int>,int,CompareNgrams> counts_t;
+		typedef map<vector<int>,int,CompareNgrams>::iterator counts_it;
+
+		typedef vector<counts_t*> refcounts_t;
+
+		size_t countNgrams(const string& line, counts_t& counts, unsigned int n);
+
+		void dump_counts(counts_t& counts) {
+			for (counts_it i = counts.begin(); i != counts.end(); ++i) {
+				cerr << "(";
+				copy(i->first.begin(), i->first.end(), ostream_iterator<int>(cerr," "));
+				cerr << ") " << i->second << ", ";
+			}
+			cerr << endl;
+		} 
+		BleuReferenceLengthStrategy _refLengthStrategy;
+		
+		// data extracted from reference files
+		refcounts_t _refcounts;
+		vector<vector<size_t> > _reflengths;
+};
+
+
+
+
+/**
+  * Implementation of position-independent word error rate. This is defined
+  * as 1 - (correct - max(0,output_length - ref_length)) / ref_length
+  * In fact, we ignore the " 1 - " so that it can be maximised.
+ **/
+class PerScorer: public StatisticsBasedScorer {
+	public:
+		PerScorer() : StatisticsBasedScorer("PER") {}
+		virtual void setReferenceFiles(const vector<string>& referenceFiles);
+		virtual void prepareStats(int sid, const string& text, ScoreStats& entry);
+
+    protected:
+        
+        virtual float calculateScore(const vector<int>& comps) ;
+		
+	private:
+        
+		//no copy
+		PerScorer(const PerScorer&);
+		PerScorer& operator=(const PerScorer&);
+				
+		// data extracted from reference files
+		vector<size_t> _reflengths;
+        vector<multiset<int> > _reftokens;
+};
+
+
+class ScorerFactory {
+
+    public:
+        vector<string> getTypes() {
+            vector<string> types;
+            types.push_back(string("BLEU"));
+            types.push_back(string("PER"));
+            return types;
+        }
+
+        Scorer* getScorer(string type) {
+            if (type == "BLEU") {
+                return new BleuScorer();
+            } else if (type == "PER") {
+                return new PerScorer();
+            } else {
+                throw runtime_error("Unknown scorer type: " + type);
+            }
+       }
+};
+
 #endif //__SCORER_H
diff --git a/mert/extractor.cpp b/mert/extractor.cpp
index 524dcff96..bb47bf0d2 100644
--- a/mert/extractor.cpp
+++ b/mert/extractor.cpp
@@ -7,8 +7,6 @@ using namespace std;
 #include "Util.h"
 
 #include "Scorer.h"
-#include "BleuScorer.h"
-#include "PerScorer.h"
 #include "Data.h"
 
 int main (int argc, char * argv[]) {
diff --git a/mert/test_scorer.cpp b/mert/test_scorer.cpp
index 8e6b70ffb..bc1ac481f 100644
--- a/mert/test_scorer.cpp
+++ b/mert/test_scorer.cpp
@@ -2,8 +2,7 @@
 #include <vector>
 
 #include "ScoreData.h"
-#include "BleuScorer.h"
-#include "PerScorer.h"
+#include "Scorer.h"
 
 using namespace std;