Fixed accuracy computing

trainer/include/TrainInfos.hpp

@@ -18,7 +18,6 @@ class TrainInfos
   std::string filename;
   int lastEpoch;
   int lastSaved;
-  int lastIndexTreated;
   std::map< std::string, std::vector<float> > trainLossesPerClassifierPerEpoch;
   std::map< std::string, std::vector<float> > devLossesPerClassifierPerEpoch;
   std::map< std::string, std::vector<float> > trainScoresPerClassifierPerEpoch;
@@ -36,7 +35,7 @@ class TrainInfos
   void saveToFilename();
   void addTrainScore(const std::string & classifier, float score);
   void addDevScore(const std::string & classifier, float score);
-  float computeScoreOnTapes(Config & c, std::vector<std::string> tapes);
+  float computeScoreOnTapes(Config & c, std::vector<std::string> tapes, int from, int to);
 
   public :
 

trainer/src/TrainInfos.cpp

@@ -143,12 +143,12 @@ void TrainInfos::addDevScore(const std::string & classifier, float score)
   devScoresPerClassifierPerEpoch[classifier].emplace_back(score);
 }
 
-float TrainInfos::computeScoreOnTapes(Config & c, std::vector<std::string> tapes)
+float TrainInfos::computeScoreOnTapes(Config & c, std::vector<std::string> tapes, int from, int to)
 {
   float res = 0.0;
 
   for (auto & tape : tapes)
-    res += c.getTape(tape).getScore(0, lastIndexTreated);
+    res += c.getTape(tape).getScore(from, to);
 
   return res / tapes.size();
 }
@@ -158,13 +158,13 @@ void TrainInfos::computeTrainScores(Config & c)
   for (auto & it : topologyPrinted)
   {
     if (it.first == "Parser")
-      addTrainScore(it.first, computeScoreOnTapes(c, {"GOV", "LABEL"}));
+      addTrainScore(it.first, computeScoreOnTapes(c, {"GOV", "LABEL"}, 0, c.getHead()));
     else if (it.first == "Tagger")
-      addTrainScore(it.first, computeScoreOnTapes(c, {"POS"}));
+      addTrainScore(it.first, computeScoreOnTapes(c, {"POS"}, 0, c.getHead()));
     else if (it.first == "Morpho")
-      addTrainScore(it.first, computeScoreOnTapes(c, {"MORPHO"}));
+      addTrainScore(it.first, computeScoreOnTapes(c, {"MORPHO"}, 0, c.getHead()));
     else if (it.first == "Lemmatizer_Rules")
-      addTrainScore(it.first, computeScoreOnTapes(c, {"LEMMA"}));
+      addTrainScore(it.first, computeScoreOnTapes(c, {"LEMMA"}, 0, c.getHead()));
     else if (split(it.first, '_')[0] == "Error")
       addTrainScore(it.first, 100.0);
     else
@@ -180,15 +180,15 @@ void TrainInfos::computeDevScores(Config & c)
   for (auto & it : topologyPrinted)
   {
     if (it.first == "Parser")
-      addDevScore(it.first, computeScoreOnTapes(c, {"GOV", "LABEL"}));
+      addDevScore(it.first, computeScoreOnTapes(c, {"GOV", "LABEL"}, 0, c.getHead()));
     else if (it.first == "Parser")
-      addDevScore(it.first, computeScoreOnTapes(c, {"GOV", "LABEL"}));
+      addDevScore(it.first, computeScoreOnTapes(c, {"GOV", "LABEL"}, 0, c.getHead()));
     else if (it.first == "Tagger")
-      addDevScore(it.first, computeScoreOnTapes(c, {"POS"}));
+      addDevScore(it.first, computeScoreOnTapes(c, {"POS"}, 0, c.getHead()));
     else if (it.first == "Morpho")
-      addDevScore(it.first, computeScoreOnTapes(c, {"MORPHO"}));
+      addDevScore(it.first, computeScoreOnTapes(c, {"MORPHO"}, 0, c.getHead()));
     else if (it.first == "Lemmatizer_Rules")
-      addDevScore(it.first, computeScoreOnTapes(c, {"LEMMA"}));
+      addDevScore(it.first, computeScoreOnTapes(c, {"LEMMA"}, 0, c.getHead()));
     else if (split(it.first, '_')[0] == "Error")
       addDevScore(it.first, 100.0);
     else
@@ -292,8 +292,3 @@ bool TrainInfos::mustSave(const std::string & classifier)
   return mustSavePerClassifierPerEpoch.count(classifier) && mustSavePerClassifierPerEpoch[classifier].back();
 }
 
-void TrainInfos::setLastIndexTreated(int index)
-{
-  lastIndexTreated = index;
-}
-

trainer/src/Trainer.cpp

@@ -192,7 +192,6 @@ void Trainer::resetAndShuffle()
 {
   tm.reset();
   trainConfig.reset();
-  TI.setLastIndexTreated(0);
 
   if(ProgramParameters::shuffleExamples)
     trainConfig.shuffle(ProgramParameters::sequenceDelimiterTape, ProgramParameters::sequenceDelimiter);
@@ -541,8 +540,6 @@ void Trainer::train()
       if (ProgramParameters::iterationSize != -1 && nbSteps >= ProgramParameters::iterationSize)
         try {prepareNextEpoch();}
         catch (EndOfTraining &) {break;}
-
-      TI.setLastIndexTreated(trainConfig.getHead());
     }
 
     if (ProgramParameters::debug)