public ConvolutionalChromosome(ConvolutionalNetwork cNet, ConvRandomParams crp, DenseRandomParams drp)
{
this.crp = crp;
this.drp = drp;
//добавить рандомные параметры
convPart = new ConvStructure(cNet.slidingWindows, cNet.convActivationIndexes, cNet.filters, cNet.convDropoutIndexes, cNet.convDropoutRates);
densePart = new DenseStructure(cNet.denseActivationIndexes, cNet.neurons, cNet.denseDropoutIndexes, cNet.denseDropoutRates);
} // конструктор для создания хромосомы из полей описательного класса ConvolutionalNetwork для последующего использования в генетических алгоритмах (например, для поиска лучшего варианта чем текущий)
///////////////////////////////////////////////////////
// new Process
public void UpdateAssessmentParams(ConvolutionalChromosome chromosome, GeneticProgramm gp, int epoch)
{
instance.Invoke(new Action(() => {
if (!instance.errDontClearChB.Checked)
{
instance.ErrorTB.Clear();
}
if (!instance.chrDontClearChB.Checked)
{
instance.chrOutTB.Clear();
}
}));
ConvolutionalNetwork tempNet = new ConvolutionalNetwork(chromosome, epoch, gp._nrp.networkName);
ConvolutionalNetwork.CreateNetworkScript(tempNet, chromosome.crp.convActivations, chromosome.drp.denseActivations);
ConvolutionalNetwork.CreateConvBatFile("temp_convolution.py", @"C:\keras\Directory\scripts\convolutional\genetic", gp._nrp.datasetPath,
gp._nrp._modelPath, gp._nrp._labelPath, gp._nrp._plotPath, gp._nrp.networkName, epoch, chromosome.indexNumber, 0);
CreateConvBatProcess();
chromosome.accuracy = instance.tempAccuracy;
chromosome.paramsCount = instance.tempParameters;
}
public static void InsertConvNetworkCode(String path, int line, ConvolutionalNetwork network, List <String> convActivations, List <String> denseActivations)
{
string loss_function = "\"categorical_crossentropy\"";
if (network.outputs == 2)
{
loss_function = "\"binary_crossentropy\"";
}
string init_lr = network.trainConstSpeed.ToString();
init_lr = init_lr.Replace(",", ".");
string callbacks_str = "";
string code = "";
for (int i = 0; i < network.filters.Count; i++)
{
if (i == 0)
{
code += "model.add(ZeroPadding2D((1, 1), input_shape = (128, 128, 3)))" + Environment.NewLine;
}
code += "model.add(Conv2D(" + network.filters[i].ToString() + ", kernel_size = (" +
network.slidingWindows[i][0].ToString() + ", " + network.slidingWindows[i][1].ToString() +
"), strides = (1, 1), padding='same', activation = '" + convActivations[network.convActivationIndexes[i]] + "'))" + Environment.NewLine;
code += "model.add(MaxPooling2D(pool_size = (2, 2), strides = (2, 2)))" + Environment.NewLine;
}
code += "model.add(Flatten())" + Environment.NewLine;
int dropoutCounter = 0;
for (int i = 0; i < network.neurons.Count; i++)
{
code += "model.add(Dense(" + network.neurons[i].ToString() + ", activation = '" +
denseActivations[network.denseActivationIndexes[i]] + "'))" + Environment.NewLine;
if (network.denseDropoutIndexes.Count > 0)
{
if (network.denseDropoutIndexes[dropoutCounter] == i)
{
code += "model.add(Dropout(" + ((float)network.denseDropoutRates[dropoutCounter] / 100).ToString() + "))" + Environment.NewLine;
if ((network.denseDropoutIndexes.Count - 1) != dropoutCounter)
{
dropoutCounter++;
}
}
}
}
if (network.outputs == 2)
{
code += "model.add(Dense(2, activation = \"softmax\"))" + Environment.NewLine;
}
else
{
code += "model.add(Dense(len(lb.classes_), activation = \"softmax\"))" + Environment.NewLine;
}
code += "INIT_LR = " + init_lr + Environment.NewLine +
"EPOCHS = " + network.networkLearningEpochs.ToString() + Environment.NewLine +
"BS = " + network.batchSize.ToString() + Environment.NewLine + Environment.NewLine;
code += "print(\"[INFO] training network...\")" + Environment.NewLine;
code += "opt = SGD(lr=INIT_LR)" + Environment.NewLine;
code += "model.compile(loss=" + loss_function + ", optimizer=opt, metrics=[\"acc\"])" + Environment.NewLine;
//code += "" + Environment.NewLine;
Support.insertLineToFile(path, line, code);
string forBinaryCode = "trainY = to_categorical(trainY)" + Environment.NewLine +
"testY = to_categorical(testY)" + Environment.NewLine;
if (network.outputs == 2)
{
Support.insertLineToFile(path, 66, forBinaryCode);
}
}
public static void CreateNetworkScript(/*String toCopyPath, String copyPath, int line,*/ ConvolutionalNetwork network, List <String> convActivations, List <String> denseActivations)
{
string fileName; // = "convolutional_v" + Properties.Settings.Default.convScriptNumber + ".py";
fileName = "temp_convolution.py";
string path = @"C:\keras\Directory\scripts\convolutional\genetic\";
//path = @"C:\keras\Directory\scripts\convolutional\genetic\elites\";
try
{
File.Copy(@"C:\keras\Directory\scripts\convolutional\convolutional.py", path + fileName, true);
}
catch (Exception ex)
{
//MessageBox.Show(ex.Message);
}
InsertConvNetworkCode(path + fileName, 75, network, convActivations, denseActivations);
//Properties.Settings.Default.convScriptNumber++;
//ConfigEditor.Config.Write("convScriptNumber", Properties.Settings.Default.convScriptNumber);
}
