Making sure optimizer have access to embeddings parameters

reading_machine/include/Classifier.hpp

@@ -13,6 +13,7 @@ class Classifier
   std::map<std::string,std::unique_ptr<TransitionSet>> transitionSets;
   std::shared_ptr<NeuralNetworkImpl> nn;
   std::unique_ptr<torch::optim::Adam> optimizer;
+  std::string optimizerType, optimizerParameters;
   std::string state;
 
   private :
@@ -27,6 +28,7 @@ class Classifier
   NeuralNetwork & getNN();
   const std::string & getName() const;
   int getNbParameters() const;
+  void resetOptimizer();
   void loadOptimizer(std::filesystem::path path);
   void saveOptimizer(std::filesystem::path path);
   torch::optim::Adam & getOptimizer();

reading_machine/src/Classifier.cpp

@@ -92,17 +92,8 @@ void Classifier::initNeuralNetwork(const std::vector<std::string> & definition)
 
   if (curIndex >= definition.size() || !util::doIfNameMatch(std::regex("(?:(?:\\s|\\t)*)(?:Optimizer :|)(?:(?:\\s|\\t)*)(.*) \\{(.*)\\}"), definition[curIndex], [&curIndex,this](auto sm)
         {
-          std::string expected = "expected '(Optimizer :) Adam {lr beta1 beta2 eps decay amsgrad}'";
-          if (sm.str(1) == "Adam")
-          {
-            auto splited = util::split(sm.str(2), ' ');
-            if (splited.size() != 6 or (splited.back() != "false" and splited.back() != "true"))
-              util::myThrow(expected);
-
-            optimizer.reset(new torch::optim::Adam(getNN()->parameters(), torch::optim::AdamOptions(std::stof(splited[0])).amsgrad(splited.back() == "true").betas({std::stof(splited[1]),std::stof(splited[2])}).eps(std::stof(splited[3])).weight_decay(std::stof(splited[4]))));
-          }
-          else
-            util::myThrow(expected);
+          optimizerType = sm.str(1);
+          optimizerParameters = sm.str(2);
         }))
     util::myThrow(fmt::format("Invalid line '{}', expected '{}'\n", curIndex < definition.size() ? definition[curIndex] : "", "(Optimizer :) Adam {lr beta1 beta2 eps decay amsgrad}"));
 }
@@ -147,3 +138,18 @@ void Classifier::initModular(const std::vector<std::string> & definition, std::s
   this->nn.reset(new ModularNetworkImpl(nbOutputsPerState, modulesDefinitions));
 }
 
+void Classifier::resetOptimizer()
+{
+  std::string expected = "expected '(Optimizer :) Adam {lr beta1 beta2 eps decay amsgrad}'";
+  if (optimizerType == "Adam")
+  {
+    auto splited = util::split(optimizerParameters, ' ');
+    if (splited.size() != 6 or (splited.back() != "false" and splited.back() != "true"))
+      util::myThrow(expected);
+ 
+    optimizer.reset(new torch::optim::Adam(getNN()->parameters(), torch::optim::AdamOptions(std::stof(splited[0])).amsgrad(splited.back() == "true").betas({std::stof(splited[1]),std::stof(splited[2])}).eps(std::stof(splited[3])).weight_decay(std::stof(splited[4]))));
+  }
+  else
+    util::myThrow(expected);
+}
+

trainer/src/MacaonTrain.cpp

@@ -167,6 +167,7 @@ int MacaonTrain::main()
   if (!computeDevScore)
     trainer.createDevDataset(devGoldConfig, debug, modelPath/"examples/dev", currentEpoch, dynamicOracleInterval);
 
+  machine.getClassifier()->resetOptimizer();
   auto optimizerCheckpoint = machinePath.parent_path() / "optimizer.pt";
   if (std::filesystem::exists(trainInfos))
     machine.getClassifier()->loadOptimizer(optimizerCheckpoint);