- added early stop of training

- fixed non-visible training data after load

git-svn-id: http://moon:8086/svn/software/trunk/projects/RBM@554 b431acfa-c32f-4a4a-93f1-934dc6c82436

Source/MainComponent.cpp

@@ -485,15 +485,15 @@ void MainComponent::buttonClicked (Button* buttonThatWasClicked)
     if (buttonThatWasClicked == trainButton)
     {
         //[UserButtonCode_trainButton] -- add your button handler code here..
-		if (rbmNormalizeDataToggleButton->getToggleState() == true)
+		if (isThreadRunning())
 		{
-			m_trainingData = Rbm::normalizeData(m_layers.data());
+			m_doStop = true;
 		}
 		else
 		{
-			m_trainingData = m_layers.data();
+			m_doStop = false;
+			startThread();
 		}
-    	startThread();
         //[/UserButtonCode_trainButton]
     }
     else if (buttonThatWasClicked == addButton)
@@ -874,11 +874,19 @@ const juce::String& MainComponent::getBaseDir()
 
 void MainComponent::run()
 {
-	m_pRbmComponentCurr->train((size_t)numEpochslabel->getText().getIntValue(), (size_t)sizeMiniBatch->getText().getIntValue());
+	m_pRbmComponentCurr->train((size_t)numEpochslabel->getText().getIntValue(), (size_t)sizeMiniBatch->getText().getIntValue(), m_doStop);
 }
 
 void MainComponent::onChanged(const LayerArray &obj)
 {
+	if (rbmNormalizeDataToggleButton->getToggleState() == true)
+	{
+		m_trainingData = Rbm::normalizeData(m_layers.data());
+	}
+	else
+	{
+		m_trainingData = m_layers.data();
+	}
 	patterSlider->setRange(0, obj.getSize()-1, 1);
 }
 

Source/MainComponent.h

@@ -90,6 +90,7 @@ private:
 	void run();
 	String m_baseDir;
 	TooltipWindow m_toolTipWindow;
+	bool m_doStop;
 	void clearTraining()
 	{
 		m_layers.clear();

Source/Rbm.cpp

@@ -98,7 +98,7 @@ MatrixXd Rbm::normalizeData(MatrixXd const &src)
 	
 }
 
-void Rbm::train(size_t numEpochs, size_t miniBatchSize, double sigmaMin)
+void Rbm::train(size_t numEpochs, size_t miniBatchSize, bool &doStop)
 {
 	size_t i;
 	size_t epoch;
@@ -124,6 +124,11 @@ void Rbm::train(size_t numEpochs, size_t miniBatchSize, double sigmaMin)
 	m_progress = 0;
 	while (trainingSizeRemain)
 	{
+		if (doStop)
+		{
+			break;
+		}
+		
 		cout << "trainingSizeRemain: " << trainingSizeRemain << endl;
 		size_t toSlice = std::min(miniBatchSize, trainingSizeRemain);
 		MatrixXd batch = __batch.block(batchRowIndex, 0, toSlice, m_w.getNumVisible());
@@ -140,6 +145,10 @@ void Rbm::train(size_t numEpochs, size_t miniBatchSize, double sigmaMin)
 
 		for (epoch=0; epoch < numEpochs; epoch++)
 		{
+			if (doStop)
+			{
+				break;
+			}
 			onProgressChanged();
 
 			// Create hidden layer base on training data

Source/Rbm.hpp

@@ -48,7 +48,7 @@ public:
 	void sampleGaussian(MatrixXd &dst, MatrixXd const &src);
 	void sampleGaussian(MatrixXd &srcDst);
 	static MatrixXd normalizeData(MatrixXd const &src);
-	void train(size_t numEpochs, size_t miniBatchSize, double sigmaMin = 0.05);
+	void train(size_t numEpochs, size_t miniBatchSize, bool &doStop);
 	double getProgress() const;
 	void toHidden(RowVectorXd &h, RowVectorXd const &v);
 	void toVisible(RowVectorXd &v, RowVectorXd const &h);