- trainingdata always contains context

- load / store training batch with context
- on load: add context part to  legacy training batches 
- removed Rbm::setBatch()

git-svn-id: http://moon:8086/svn/software/trunk/projects/Rbm@790 b431acfa-c32f-4a4a-93f1-934dc6c82436

source/MainComponent.cpp

@@ -532,8 +532,10 @@ void MainComponent::buttonClicked (Button* buttonThatWasClicked)
     else if (buttonThatWasClicked == addButton)
     {
         //[UserButtonCode_addButton] -- add your button handler code here..
+		Layer *layer = m_stack->getLayer(0);
 		RbmComponent *pComp = static_cast<RbmComponent*>(m_stack->getLayer(0));
-		addTraining(pComp->DrawVisibleTrain->getData());
+		addTraining(arma::join_rows(pComp->DrawVisibleTrain->getData(), pComp->DrawContextTrain->getData()));
+		
         //[/UserButtonCode_addButton]
     }
     else if (buttonThatWasClicked == ShakeButton)
@@ -899,8 +901,8 @@ const juce::String& MainComponent::getBaseDir()
 void MainComponent::run()
 {
     trainButton->setButtonText (TRANS("Stop"));
-	m_pLayer->setBatch(m_stack->trainingBatch());
-	m_pLayer->train(this);
+	m_pLayer->calcContextBatch(m_stack->trainingBatch());
+	m_pLayer->train(m_stack->trainingBatch(), this);
     trainButton->setButtonText (TRANS("Train"));
 }
 

source/MainComponent.hpp

@@ -126,18 +126,7 @@ private:
 
 	const arma::mat trainingAt(size_t index)
 	{
-		if (m_pLayer->context().n_cols > 0)
-		{
-			if (m_pLayer->context().n_rows != m_stack->trainingBatch().n_rows)
-			{
-				m_pLayer->setBatch(m_stack->trainingBatch());
-			}
-			return arma::join_rows(m_stack->trainingBatch().row(index), m_pLayer->context().row(index));
-		}
-		else
-		{	
-			return m_stack->trainingBatch().row(index);
-		}
+		return m_stack->trainingBatch().row(index);
 	}
 	
 	void updateControls();

source/Rbm.cpp

@@ -116,10 +116,9 @@ void Rbm::contrastiveDivergence(arma::mat const &v_states, arma::mat &dw, arma::
 	dbhv -= sum(h_probs, 0);
 }
 
-void Rbm::train(IListener* pListener)
+void Rbm::train(arma::mat const &batch, IListener* pListener)
 {
 	Status status;
-	arma::mat const &batch = m_batch;
 	double dProgress = 100.0/(batch.n_rows*m_params.numEpochs);
 	double progress = 0;
 	int lastProgress = -100;

source/Rbm.hpp

@@ -126,12 +126,7 @@ public:
 		m_bv.submat(0, 0, bv.n_rows-1, bv.n_cols-1) = bv;
 	}
 	
-	void setBatch(arma::mat const &batch)
-	{
-		m_batch = batch;
-	}
-	
-	void train(IListener *pListener=nullptr);
+	void train(arma::mat const &batch, IListener *pListener=nullptr);
 	
 	static arma::mat normalize(const arma::mat &hidden);
 	const arma::mat& whv() const;
@@ -196,7 +191,6 @@ protected:
 private:
 	noise_gen_t m_noise;
 	arma::mat m_whv;
-	arma::mat m_batch;
 	
 };
 

source/Stack.cpp

@@ -204,8 +204,8 @@ void Stack::train(Rbm::IListener* pListener)
 	while(pLayer)
 	{
 		std::cout << m_name << ": " << " Training of layer " << std::to_string(pLayer->id()) << std::endl;
-		pLayer->setBatch(m_trainingBatch);
-		pLayer->train(pListener);
+		pLayer->calcContextBatch(m_trainingBatch);
+		pLayer->train(m_trainingBatch, pListener);
 		pLayer = pLayer->next;
 	}
 }
@@ -245,6 +245,16 @@ size_t Stack::loadTrainingBatch(bool doNormalize)
 		}
 		std::cout << "Loaded " << m_trainingBatch.n_rows << " training samples\n";
 	}
+	
+	// Migrate context part to training data
+	size_t numContext = getLayer(0)->context().n_cols;
+	size_t numTraining = m_trainingBatch.n_rows;
+	if (numContext > 0 and m_trainingBatch.n_cols != getLayer(0)->numVisible())
+	{
+		arma::mat training_with_ctx = arma::join_rows(m_trainingBatch, arma::zeros(numTraining, numContext));
+		m_trainingBatch = training_with_ctx;
+	}
+
 	return m_trainingBatch.n_rows;
 }
 

source/Stack.hpp

@@ -65,7 +65,6 @@ private:
 	std::string m_name;
 	Layer *m_pLayers;
 	arma::mat m_trainingBatch;
-	
 };
 
 

source/main.cpp

@@ -81,8 +81,6 @@ int main()
 	
 	Stack stack(".", project);
 
-	stack.loadTrainingBatch();
-
 #if CREATE_TEST
 	stack.addTraining(stack.trainingBatch().row(1));
 	printf("There are %d training samples\n", (int)stack.trainingBatch().n_rows);
@@ -119,6 +117,10 @@ int main()
 	
 	// Load weights
 	stack.loadWeights();
+	
+	// Load training
+	stack.loadTrainingBatch();
+
 #endif	
 	
 #if TRAIN_TEST
@@ -132,6 +134,8 @@ int main()
 #endif
 	Layer *layer = stack.getLayer(0);
 	arma::mat v = stack.trainingBatch();
+	v.print("t");
+	layer->calcContextBatch(v);
 	arma::mat h = layer->toHiddenProbs(v);
 	arma::mat r = layer->toVisibleProbs(h);