private void button1_Click_1(object sender, EventArgs e)
{
NeuralNetwork nn = new NeuralNetwork(new int[] { 784, 20, 100, 10, 10 });
Log("Loading testing set..");
List<double[]> testQuestions = MNISTTools.AdjustImages(MNISTTools.ReadImages("mnist/t10k-images.idx3-ubyte")).GetRange(0, 1000);
List<double[]> testAnswers = MNISTTools.AdjustLabels(MNISTTools.ReadLabels("mnist/t10k-labels.idx1-ubyte")).GetRange(0, 1000);
Log("Testing network..");
double correctAnswers = TestNetwork(testQuestions, testAnswers, nn);
Log("Correct answers: " + correctAnswers);
nn.Save("lst.bin");
NeuralNetwork nn2 = new NeuralNetwork("lst.bin");
Log("Testing network..");
correctAnswers = TestNetwork(testQuestions, testAnswers, nn2);
Log("Correct answers: " + correctAnswers);
}
private void button2_Click(object sender, EventArgs e)
{
Log("Loading training data...");
List<double[]> trainQuestions = MNISTTools.AdjustImages(MNISTTools.ReadImages("mnist/train-images.idx3-ubyte"));
List<double[]> trainAnswers = MNISTTools.AdjustLabels(MNISTTools.ReadLabels("mnist/train-labels.idx1-ubyte"));
Log("Loading testing set..");
List<double[]> testQuestions = MNISTTools.AdjustImages(MNISTTools.ReadImages("mnist/t10k-images.idx3-ubyte"));
List<double[]> testAnswers = MNISTTools.AdjustLabels(MNISTTools.ReadLabels("mnist/t10k-labels.idx1-ubyte"));
Log("Creating network..");
// Parsing configuration
string[] hiddenLayersConf = textBox_configuration.Text.Split(' ');
int[] configuration = new int[hiddenLayersConf.Length + 2];
configuration[0] = trainQuestions[0].Length;
configuration[configuration.Length - 1] = trainAnswers[0].Length;
for (int i = 0; i < hiddenLayersConf.Length; i++)
{
configuration[i + 1] = Convert.ToInt32(hiddenLayersConf[i]);
}
// Parsing speed coefficients
double kBegin = double.Parse(textBox_kSettings.Text.Split(' ')[0], CultureInfo.InvariantCulture);
double kDelta = double.Parse(textBox_kSettings.Text.Split(' ')[1], CultureInfo.InvariantCulture);
double kEnd = double.Parse(textBox_kSettings.Text.Split(' ')[2], CultureInfo.InvariantCulture);
Log("Speed: kBegin: " + kBegin + " kDelta: " + kDelta + " kEnd: " + kEnd);
Log("Configuration: [" + string.Join(" ", configuration) + "]");
NeuralNetwork nn = new NeuralNetwork(configuration);
double k = kBegin;
Log("Teaching network..");
while (k >= kEnd)
{
TeachSet(trainQuestions, trainAnswers, k, nn);
double globalError = nn.GlobalError(testQuestions, testAnswers);
Log("Global error: " + globalError + " k: " + k);
k -= kDelta;
}
Log("Testing network..");
double correctAnswers = TestNetwork(trainQuestions, trainAnswers, nn);
Log("Correct answers: " + correctAnswers);
Log("Saving..");
string fileName = textBox_outFileName.Text;
nn.Save(fileName + ".bin");
File.WriteAllLines(fileName + ".txt", new string[] { richTextBox1.Text });
Log("ok");
}
private double TestNetwork(List<double[]> trainQuestions, List<double[]> trainAnswers, NeuralNetwork nn)
{
int correctAnswers = 0;
for (int i = 0; i < trainQuestions.Count; i++)
{
double[] nnOut = nn.Classify(trainQuestions[i]);
int maxInd = 0;
for (int j = 0; j < nnOut.Length; j++)
{
if (nnOut[j] > nnOut[maxInd])
{
maxInd = j;
}
}
if (trainAnswers[i][maxInd] == 1)
{
correctAnswers++;
}
}
return (double)correctAnswers / trainQuestions.Count;
}
private void TeachSet(List<double[]> inputs, List<double[]> outputs, double k, NeuralNetwork nn)
{
for (int i = 0; i < inputs.Count; i++)
{
nn.Learn(inputs[i], outputs[i], k);
}
}