Added new option for training

maca_common/include/ProgramParameters.hpp

@@ -76,6 +76,7 @@ struct ProgramParameters
   static bool randomDebug;
   static float randomDebugProbability;
   static bool alwaysSave;
+  static bool noNeuralNetwork;
 
   private :
 

maca_common/src/ProgramParameters.cpp

@@ -70,3 +70,4 @@ bool ProgramParameters::devLoss;
 bool ProgramParameters::randomDebug;
 float ProgramParameters::randomDebugProbability;
 bool ProgramParameters::alwaysSave;
+bool ProgramParameters::noNeuralNetwork;

maca_common/src/programOptionsTemplates.cpp

@@ -42,6 +42,7 @@ po::options_description getTrainOptionsDescription()
     ("help,h", "Produce this help message")
     ("debug,d", "Print infos on stderr")
     ("alwaysSave", "Save the model at every iteration")
+    ("noNeuralNetwork", "Don't use any neural network, useful to speed up debug")
     ("randomDebug", "Print infos on stderr with a probability of randomDebugProbability")
     ("randomDebugProbability", po::value<float>()->default_value(0.001),
       "Probability that debug infos will be printed")
@@ -270,6 +271,7 @@ void loadTrainProgramParameters(int argc, char * argv[])
   ProgramParameters::mcdName = vm["mcd"].as<std::string>();
   ProgramParameters::debug = vm.count("debug") == 0 ? false : true;
   ProgramParameters::alwaysSave = vm.count("alwaysSave") == 0 ? false : true;
+  ProgramParameters::noNeuralNetwork = vm.count("noNeuralNetwork") == 0 ? false : true;
   ProgramParameters::randomDebug = vm.count("randomDebug") == 0 ? false : true;
   ProgramParameters::printEntropy = vm.count("printEntropy") == 0 ? false : true;
   ProgramParameters::printTime = vm.count("printTime") == 0 ? false : true;

transition_machine/src/Classifier.cpp

@@ -94,6 +94,14 @@ Classifier::WeightedActions Classifier::weightActions(Config & config)
 {
   WeightedActions result;
 
+  if (ProgramParameters::noNeuralNetwork)
+  {
+    for (unsigned int i = 0; i < as->actions.size(); i++)
+      result.emplace_back(as->actions[i].appliable(config), std::pair<float, std::string>(1.0, as->actions[i].name));
+
+    return result;
+  }
+
   if(type == Type::Prediction)
   {
     initClassifier(config);
@@ -269,17 +277,26 @@ std::string Classifier::getDefaultAction() const
 
 float Classifier::trainOnExample(Config & config, int gold)
 {
+  if (ProgramParameters::noNeuralNetwork)
+    return 0.0;
+
   auto & fd = fm->getFeatureDescription(config);
   return nn->update(fd, gold);
 }
 
 float Classifier::trainOnExample(FeatureModel::FeatureDescription & fd, int gold)
 {
+  if (ProgramParameters::noNeuralNetwork)
+    return 0.0;
+
   return nn->update(fd, gold);
 }
 
 float Classifier::getLoss(Config & config, int gold)
 {
+  if (ProgramParameters::noNeuralNetwork)
+    return 0.0;
+
   auto & fd = fm->getFeatureDescription(config);
   return nn->getLoss(fd, gold);
 }