fix future and total cost in multimodel(counts). (was broken since merge of branch weight-new in May)

4 files changed, 40 insertions, 16 deletions

diff --git a/moses/ScoreComponentCollection.cpp b/moses/ScoreComponentCollection.cpp
index d6ddaec9f..97ae2684d 100644
--- a/moses/ScoreComponentCollection.cpp
+++ b/moses/ScoreComponentCollection.cpp
@@ -185,6 +185,19 @@ void ScoreComponentCollection::Assign(const FeatureFunction* sp, const string li
   }
 }
 
+void ScoreComponentCollection::InvertDenseFeatures(const FeatureFunction* sp)
+{
+
+  Scores old_scores = GetScoresForProducer(sp);
+  Scores new_scores(old_scores.size());
+
+  for (size_t i = 0; i != old_scores.size(); ++i) {
+    new_scores[i] = -old_scores[i];
+  }
+
+  Assign(sp, new_scores);
+}
+
 void ScoreComponentCollection::ZeroDenseFeatures(const FeatureFunction* sp)
 {
   size_t numScores = sp->GetNumScoreComponents();
diff --git a/moses/ScoreComponentCollection.h b/moses/ScoreComponentCollection.h
index a35cfde71..a7c4eb2b7 100644
--- a/moses/ScoreComponentCollection.h
+++ b/moses/ScoreComponentCollection.h
@@ -335,6 +335,7 @@ public:
   float GetWeightedScore() const;
 
   void ZeroDenseFeatures(const FeatureFunction* sp);
+  void InvertDenseFeatures(const FeatureFunction* sp);
   void L1Normalise();
   float GetL1Norm() const;
   float GetL2Norm() const;
diff --git a/moses/TranslationModel/PhraseDictionaryMultiModel.cpp b/moses/TranslationModel/PhraseDictionaryMultiModel.cpp
index d9c53d5f9..77b611335 100644
--- a/moses/TranslationModel/PhraseDictionaryMultiModel.cpp
+++ b/moses/TranslationModel/PhraseDictionaryMultiModel.cpp
@@ -136,19 +136,19 @@ void PhraseDictionaryMultiModel::CollectSufficientStatistics(const Phrase& src,
 
           multiModelStatistics * statistics = new multiModelStatistics;
           statistics->targetPhrase = new TargetPhrase(*targetPhrase); //make a copy so that we don't overwrite the original phrase table info
-
-          // zero out scores from original phrase table
-          statistics->targetPhrase->GetScoreBreakdown().ZeroDenseFeatures(&pd);
-
-          Scores scoreVector(m_numScoreComponents);
           statistics->p.resize(m_numScoreComponents);
           for(size_t j = 0; j < m_numScoreComponents; ++j) {
             statistics->p[j].resize(m_numModels);
-            scoreVector[j] = -raw_scores[j];
           }
 
-          statistics->targetPhrase->GetScoreBreakdown().Assign(this, scoreVector); // set scores to 0
-          statistics->targetPhrase->Evaluate(src, GetFeaturesToApply());
+          //correct future cost estimates and total score
+          statistics->targetPhrase->GetScoreBreakdown().InvertDenseFeatures(&pd);
+          vector<FeatureFunction*> pd_feature;
+          pd_feature.push_back(m_pd[i]);
+          const vector<FeatureFunction*> pd_feature_const(pd_feature);
+          statistics->targetPhrase->Evaluate(src, pd_feature_const);
+          // zero out scores from original phrase table
+          statistics->targetPhrase->GetScoreBreakdown().ZeroDenseFeatures(&pd);
 
           (*allStats)[targetString] = statistics;
 
@@ -183,7 +183,12 @@ TargetPhraseCollection* PhraseDictionaryMultiModel::CreateTargetPhraseCollection
     scoreVector[m_numScoreComponents-1] = 1.0;
 
     statistics->targetPhrase->GetScoreBreakdown().Assign(this, scoreVector);
-    statistics->targetPhrase->Evaluate(src, GetFeaturesToApply());
+
+    //correct future cost estimates and total score
+    vector<FeatureFunction*> pd_feature;
+    pd_feature.push_back(const_cast<PhraseDictionaryMultiModel*>(this));
+    const vector<FeatureFunction*> pd_feature_const(pd_feature);
+    statistics->targetPhrase->Evaluate(src, pd_feature_const);
 
     ret->Add(new TargetPhrase(*statistics->targetPhrase));
   }
diff --git a/moses/TranslationModel/PhraseDictionaryMultiModelCounts.cpp b/moses/TranslationModel/PhraseDictionaryMultiModelCounts.cpp
index 4af3060ea..7b9320827 100644
--- a/moses/TranslationModel/PhraseDictionaryMultiModelCounts.cpp
+++ b/moses/TranslationModel/PhraseDictionaryMultiModelCounts.cpp
@@ -183,17 +183,17 @@ void PhraseDictionaryMultiModelCounts::CollectSufficientStatistics(const Phrase&
           multiModelCountsStatistics * statistics = new multiModelCountsStatistics;
           statistics->targetPhrase = new TargetPhrase(*targetPhrase); //make a copy so that we don't overwrite the original phrase table info
 
+          //correct future cost estimates and total score
+          statistics->targetPhrase->GetScoreBreakdown().InvertDenseFeatures(&pd);
+          vector<FeatureFunction*> pd_feature;
+          pd_feature.push_back(m_pd[i]);
+          const vector<FeatureFunction*> pd_feature_const(pd_feature);
+          statistics->targetPhrase->Evaluate(src, pd_feature_const);
           // zero out scores from original phrase table
           statistics->targetPhrase->GetScoreBreakdown().ZeroDenseFeatures(&pd);
 
           statistics->fst.resize(m_numModels);
           statistics->ft.resize(m_numModels);
-          Scores scoreVector(5);
-          scoreVector[0] = -raw_scores[0];
-          scoreVector[1] = -raw_scores[1];
-          scoreVector[2] = -raw_scores[2];
-          statistics->targetPhrase->GetScoreBreakdown().Assign(this, scoreVector); // set scores to 0
-          statistics->targetPhrase->Evaluate(src, GetFeaturesToApply());
 
           (*allStats)[targetString] = statistics;
 
@@ -246,7 +246,12 @@ TargetPhraseCollection* PhraseDictionaryMultiModelCounts::CreateTargetPhraseColl
       scoreVector[4] = FloorScore(TransformScore(2.718));
 
       statistics->targetPhrase->GetScoreBreakdown().Assign(this, scoreVector);
-      statistics->targetPhrase->Evaluate(src, GetFeaturesToApply());
+
+      //correct future cost estimates and total score
+      vector<FeatureFunction*> pd_feature;
+      pd_feature.push_back(const_cast<PhraseDictionaryMultiModelCounts*>(this));
+      const vector<FeatureFunction*> pd_feature_const(pd_feature);
+      statistics->targetPhrase->Evaluate(src, pd_feature_const);
     } catch (AlignmentException& e) {
       continue;
     }