void EvaluateClassifier(PrinicpleComponentAnalysis PCA)
{
string[] classes = { "G", "L", "R", "Y", "Z" };
int testPatterns = 11;
double correctSamples = 0;
int classifiedPattern;
for (int i = 0; i < classes.Length; i++)
{
for (int j = 0; j < testPatterns; j++)
{
string path = "img/" + classes[i] + "/Test/" + (j + 45) + ".txt";
if (PCA != null)
{
classifiedPattern = ocrApp.ClassifyReducedData(path);
}
else
{
classifiedPattern = ocrApp.Classify(path);
}
if (classifiedPattern == i)
{
correctSamples++;
}
}
}
accuracy = correctSamples / (testPatterns * classes.Length);
misMatchingSamples = testPatterns * classes.Length - correctSamples;
}
private void trainButton_Click(object sender, EventArgs e)
{
if (IsDataComplete() == false)
{
return;
}
if (activationFunctionCmbx.SelectedItem.ToString() == "Sigmoid")
{
TransferFunction.slope = Double.Parse(valATxt.Text);
}
PrinicpleComponentAnalysis PCA = null;
if (pcaCheckBox.Checked == true)
{
int num_Components = Int32.Parse(numPCTxtBox.Text);
double pcaAlpha = Double.Parse(pcaAlphaTxtBox.Text);
TransferFunction transferFunction = new TransferFunction(TransferFunction.Type.Linear);
PCA = new PrinicpleComponentAnalysis(pcaAlpha, transferFunction, num_Components);
}
Train(PCA);
EvaluateClassifier(PCA);
output = new OutputForm(this);
output.Show();
}
void Train(PrinicpleComponentAnalysis PCA)
{
int num_HiddenLayers = Int32.Parse(hiddenLayerTxt.Text);
double alpha = Double.Parse(learningRateTxt.Text);
double threshold = Double.Parse(errorThresholdTxt.Text);
int[] layers = new int[num_HiddenLayers + 2];
int num_Classes = 5;
int num_TrainingPerClass = 44;
if (PCA != null)
{
layers[0] = PCA.num_Components;
}
else
{
layers[0] = 30 * 30;
}
layers[num_HiddenLayers + 1] = num_Classes;
for (int i = 0; i < num_HiddenLayers; i++)
{
if (neuronsPerLayerGridView.Rows[i].Cells[0].Value == null)
{
MessageBox.Show("Please Enter Number of Neurons Per Each Layer!");
return;
}
layers[i + 1] = Int32.Parse(neuronsPerLayerGridView.Rows[i].Cells[0].Value.ToString());
}
ocrApp = new OCR(layers, num_Classes, alpha, threshold, mainTransferFunction);
if (PCA != null)
{
int num_Epochs = Int32.Parse(numEpochsTxtBox.Text);
ocrApp.SetPCA(PCA, num_Epochs);
ocrApp.ReduceData(num_TrainingPerClass);
meanSquareError = ocrApp.LearnOnReducedData(num_TrainingPerClass);
}
else
{
meanSquareError = ocrApp.Learn(num_TrainingPerClass);
}
}
public void SetPCA(PrinicpleComponentAnalysis PCA, int num_Epochs)
{
this.PCA = PCA;
this.num_Epochs = num_Epochs;
}
