private static void SaveOutputNet(NeuralNetwork net, string outputFile)
{
StreamWriter sw = new StreamWriter(outputFile);
sw.Write(net);
sw.Close();
}
private static double[][] Test(NeuralNetwork net, double[][] data)
{
double[][] result = new double[data.Length][];
for (int i = 0; i < data.Length; i++)
{
result[i] = net.Output(data[i]);
}
return result;
}
static void Main(string[] args)
{
bool learn = false;
bool cp_learn = false;
String learnFilename = "";
bool test = false;
String testFilename = "";
bool useNeighbourhood = true;
bool randomWeights = false;
double[] randomWeightsLimits = null;
bool saveOutputNet = false;
String outputFile = "";
bool input = false;
String inputFilename = "";
bool bplearn = false;
bool usebias = true;
bool wh = false;
int dimension = 2;
// Learn params - default values
// 1. Kohonen params
int phases = 4;
// 2. Counter Propagation params
double ni = 0.1;
int lengthOfPhase = 1000;
double divisor = 10;
int[] neuronCounts = new int[3] { 3, 2, 1 };
for (int i = 0; i < args.Length; i++)
{
if (args[i] == "-i")
{
input = true;
inputFilename = args[i + 1];
i++;
}
if (args[i] == "-nb")
{
usebias = false;
}
if (args[i] == "-bpl")
{
bplearn = true;
learnFilename = args[i + 1];
i++;
}
if (args[i] == "-l" || args[i] == "--learn")
{
learn = true;
learnFilename = args[i + 1];
i += 2;
}
if (args[i] == "-phases")
{
phases = int.Parse(args[i + 1]);
i += 2;
}
if (args[i] == "-cpl" || args[i] == "--counter-propagation-learn")
{
cp_learn = true;
learnFilename = args[i + 1];
i++;
}
if (args[i] == "-ni")
{
ni = double.Parse(args[i + 1]);
i++;
}
if (args[i] == "-length")
{
lengthOfPhase = int.Parse(args[i + 1]);
i++;
}
if (args[i] == "-divisor")
{
divisor = double.Parse(args[i + 1]);
i++;
}
if (args[i] == "-wh")
{
wh = true;
}
if (args[i] == "-nn" || args[i] == "--no-neighbourhood")
{
useNeighbourhood = false;
}
if (args[i] == "-rw" || args[i] == "--random-weights")
{
randomWeights = true;
string limits = args[i + 1];
i++;
string[] lim = limits.Split(new char[1] { ';' }, 2);
randomWeightsLimits = new double[2];
randomWeightsLimits[0] = double.Parse(lim[0]);
randomWeightsLimits[1] = double.Parse(lim[1]);
}
if (args[i] == "-o")
{
saveOutputNet = true;
outputFile = args[i + 1];
i++;
}
if (args[i] == "-t")
{
test = true;
testFilename = args[i + 1];
i++;
}
if (args[i] == "-d")
{
dimension = int.Parse(args[i + 1]);
i++;
}
if (args[i] == "-h")
{
Console.WriteLine(File.ReadAllText("Help.txt"));
return;
}
if (args[i] == "-neurons")
{
neuronCounts = ParseNeuronCounts(args[i + 1]);
i++;
}
}
/*Console.WriteLine("wyjściowo: phases = " + phases
+ " / ni = " + ni
+ " / lengthOfPhase= " + lengthOfPhase
+ " / divisor= " + divisor);*/
NeuralNetwork net = null;
if (input)
{
net = XML.XMLNetworkCreator.Create(inputFilename);
}
if (learn)
{
double[][] data = ReadData(learnFilename);
int neuralCount = data.Length;
int inputSize = data[0].Length;
KohonenLayer kohonenLayer = new KohonenLayer(inputSize, neuralCount, randomWeights, randomWeightsLimits, dimension);
net = new NeuralNetwork(new AbstractNeuralLayer[1] { kohonenLayer });
kohonenLayer.Learn(data, 10000, useNeighbourhood, phases);
}
if (cp_learn)
{
//dane wejściowe dla warstwy Kohonena + wyjściowe dla warstwy 2giej (output programu)
List<double[][]> data = ReadInputAndAnswersData(learnFilename);
double[][] kohonenInput = data[0];
double[][] expectedOutputForSecondLayer = data[1];
int inputSize = kohonenInput[0].Length;
int neuralCount = kohonenInput.Length; // == expectedOutputForSecondLayer.Length
KohonenLayer kohonenLayer = new KohonenLayer(inputSize, neuralCount, randomWeights, randomWeightsLimits, dimension);
inputSize = neuralCount;
neuralCount = expectedOutputForSecondLayer[0].Length;
TransitionFunction tf = TransitionFunction.Sigmoid;
if (wh)
{
tf = TransitionFunction.Linear;
}
NeuralLayer secondLayer = new NeuralLayer(inputSize, neuralCount, randomWeights, randomWeightsLimits, tf, true);
net = new NeuralNetwork(new AbstractNeuralLayer[2] { kohonenLayer, secondLayer });
net.CounterPropagationLearn(kohonenInput, useNeighbourhood,
expectedOutputForSecondLayer, ni, divisor,
phases, lengthOfPhase);
}
if (bplearn)
{
List<double[][]> data = ReadInputAndAnswersData(learnFilename);
double[][] inputData = data[0];
double[][] expectedOutput = data[1];
int inputSize = inputData[0].Length;
net = new NeuralNetwork(inputSize, neuronCounts, randomWeightsLimits, TransitionFunction.Sigmoid, usebias);
net.BackPropagationLearn(inputData, expectedOutput, lengthOfPhase, ni, usebias);
}
if (saveOutputNet)
{
SaveOutputNet(net, outputFile);
}
if (test)
{
double[][] data = ReadData(testFilename);
double[][] result = Test(net, data);
PrintResult(result);
}
Console.ReadKey();
}