/// <summary>
/// Construct a BackpropagationLayer object that corresponds to a specific neuron layer.
/// </summary>
/// <param name="backpropagation">The back propagation training object.</param>
/// <param name="layer">The layer that this object corresponds to.</param>
public BackpropagationLayer(Backpropagation backpropagation,
FeedforwardLayer layer)
{
this.backpropagation = backpropagation;
this.layer = layer;
int neuronCount = layer.NeuronCount;
this.error = new double[neuronCount];
this.errorDelta = new double[neuronCount];
if (layer.Next != null)
{
this.accMatrixDelta = new Matrix.Matrix(layer.NeuronCount + 1, layer.Next.NeuronCount);
this.matrixDelta = new Matrix.Matrix(layer.NeuronCount + 1, layer.Next.NeuronCount);
this.biasRow = layer.NeuronCount;
}
}
/// <summary>
/// Construct a BackpropagationLayer object that corresponds to a specific neuron layer.
/// </summary>
/// <param name="backpropagation">The back propagation training object.</param>
/// <param name="layer">The layer that this object corresponds to.</param>
public BackpropagationLayer(Backpropagation backpropagation,
FeedforwardLayer layer)
{
this.backpropagation = backpropagation;
this.layer = layer;
int neuronCount = layer.NeuronCount;
this.error = new double[neuronCount];
this.errorDelta = new double[neuronCount];
if (layer.Next != null)
{
this.accMatrixDelta = new Matrix.Matrix(layer.NeuronCount + 1, layer.Next.NeuronCount);
this.matrixDelta = new Matrix.Matrix(layer.NeuronCount + 1, layer.Next.NeuronCount);
this.biasRow = layer.NeuronCount;
}
}
public void ThreadProc()
{
int update = 0;
Train train = new Backpropagation(this.network, PruneSelectiveForm.XOR_INPUT,
this.obtainIdeal(), 0.7, 0.9);
int max = 10000;
for (int i = 0; i < max; i++)
{
train.Iteration();
update++;
if (update == 100)
{
SetText("Cycles Left:" + (max - i) + ",Error:"
+ train.Error);
update = 0;
}
}
}
private void TrainNetworkBackpropBackprop()
{
HeatonResearchNeural.Feedforward.Train.Train train = new Backpropagation(this.network, this.input,
this.ideal, 0.7, 0.5);
int epoch = 1;
do
{
train.Iteration();
Console.WriteLine("Backprop:Iteration #" + epoch + " Error:"
+ train.Error);
epoch++;
} while ((train.Error > Config.ACCEPTABLE_ERROR));
}
/// <summary>
/// Create, train and use a neural network for XOR.
/// </summary>
/// <param name="args">Not used</param>
static void Main(string[] args)
{
/*XOR_INPUT = new double[256][];
XOR_IDEAL = new double[256][];
String[] lines = File.ReadAllLines("input.txt");
for (int i = 0; i < lines.Length; i++)
{
String str = lines[i];
double[] input = new double[8];
for (int j = 0; j < 8; j++)
{
input[j] = Double.Parse(str[j].ToString());
}
XOR_INPUT[i] = input;
}
lines = File.ReadAllLines("output.txt");
for (int i = 0; i < lines.Length; i++)
{
String str = lines[i];
double[] output = new double[1];
output[0] = Double.Parse(str);
XOR_IDEAL[i] = output;
}
*/
FeedforwardNetwork network = new FeedforwardNetwork();
network.AddLayer(new FeedforwardLayer(4));
network.AddLayer(new FeedforwardLayer(12));
network.AddLayer(new FeedforwardLayer(4));
network.Reset();
// train the neural network
Train train = new Backpropagation(network, XOR_INPUT, XOR_IDEAL,
0.2, 0.95);
int epoch = 1;
Stopwatch watch = new Stopwatch();
watch.Start();
do
{
train.Iteration();
Console.WriteLine("Epoch #" + epoch + " Error:" + train.Error);
epoch++;
} while ((epoch < 40000) && (train.Error > 0.001));
watch.Stop();
Console.WriteLine(watch.ElapsedMilliseconds / 1000.0);
// test the neural network
#if DEBUG
Console.WriteLine("Neural Network Results:");
for (int i = 0; i < XOR_IDEAL.Length; i++)
{
double[] actual = network.ComputeOutputs(XOR_INPUT[i]);
Console.WriteLine(XOR_INPUT[i][0] + "," + XOR_INPUT[i][1] + "," + XOR_INPUT[i][2] + "," + XOR_INPUT[i][3]
+ ", actual=" + actual[0] + "," + actual[1] + "," + actual[2] + "," + actual[3] + ",ideal=" + XOR_IDEAL[i][0] + "," + XOR_IDEAL[i][1] + "," + XOR_IDEAL[i][2] + "," + XOR_IDEAL[i][3]);
}
#endif
FeedforwardLayer inputLayer = network.InputLayer;
Console.WriteLine("Input layer matrix:");
Matrix layerMatrix = inputLayer.LayerMatrix;
using (StreamWriter file = new StreamWriter("weights.txt"))
{
for (int i = 0; i < layerMatrix.Rows; i++)
{
for (int j = 0; j < layerMatrix.Cols; j++)
{
Console.Write(Math.Round(layerMatrix[i, j], 3) + " ");
file.Write(layerMatrix[i, j]);
if (j != layerMatrix.Cols - 1)
{
file.Write(" ");
}
}
Console.WriteLine();
file.WriteLine();
}
file.WriteLine();
FeedforwardLayer hiddenLayer = network.HiddenLayers.ToList()[0];
Console.WriteLine("Hidden layer matrix:");
layerMatrix = hiddenLayer.LayerMatrix;
for (int i = 0; i < layerMatrix.Rows; i++)
{
for (int j = 0; j < layerMatrix.Cols; j++)
{
Console.Write(Math.Round(layerMatrix[i, j], 3) + " ");
file.Write(layerMatrix[i, j]);
if (j != layerMatrix.Cols - 1)
{
file.Write(" ");
}
}
Console.WriteLine();
file.WriteLine();
}
}
#if SHOW_MATRIX
FeedforwardLayer outputLayer = network.OutputLayer;
Console.WriteLine("Output layer matrix:");
layerMatrix = outputLayer.LayerMatrix;
for (int i = 0; i < layerMatrix.Rows; i++)
{
for (int j = 0; j < layerMatrix.Cols; j++)
{
Console.Write(layerMatrix[i, j] + " ");
}
Console.WriteLine();
}
foreach (FeedforwardLayer feedforwardLayer in network.Layers)
{
Matrix layerMatrix = feedforwardLayer.LayerMatrix;
Console.WriteLine(feedforwardLayer.Next);
for (int i = 0; i < layerMatrix.Rows; i++)
{
for (int j = 0; j < layerMatrix.Cols; j++)
{
Console.Write(layerMatrix[i, j] + " ");
}
Console.WriteLine();
}
}
#endif
Console.ReadKey();
}
private void Entrenar(int red, int pre)
{
string[] patrones = new string[ejemplos];
switch (red)
{
case 1:
for (int i = 0; i < ejemplos; i++)
{
if (pre == 0)
{
patrones[i] = pre1[i, 1];
}
else
{
patrones[i] = pos1[i, 1];
}
}
break;
case 2:
for (int i = 0; i < ejemplos; i++)
{
if (pre == 0)
{
patrones[i] = pre2[i, 1];
}
else
{
patrones[i] = pos2[i, 1];
}
}
break;
case 3:
for (int i = 0; i < ejemplos; i++)
{
if (pre == 0)
{
patrones[i] = pre3[i, 1];
}
else
{
patrones[i] = pos3[i, 1];
}
}
break;
case 4:
for (int i = 0; i < ejemplos; i++)
{
if (pre == 0)
{
patrones[i] = pre4[i, 1];
}
else
{
patrones[i] = pos4[i, 1];
}
}
break;
case 5:
for (int i = 0; i < ejemplos; i++)
{
if (pre == 0)
{
patrones[i] = pre5[i, 1];
}
else
{
patrones[i] = pos5[i, 1];
}
}
break;
}
Double[][] entrada = new double[ejemplos + 3][];
double[] matriz = new double[30];
double[] matriz2 = new double[30];
double[] matriz3 = new double[30];
//Creamos las matrices de los patrones
for (int i = 0; i < ejemplos; i++)
{
double[] matrix = new double[30];
for (int o = 0; o < 30; o++)
{
matrix[o] = double.Parse(patrones[i][o].ToString());
}
entrada[i] = matrix;
}
//Creamos 3 matrices orientativas
bool control = true;
for (int i = 0; i < 30; i++)
{
matriz3[i] = 0;
if (control)
{
matriz[i] = 1;
matriz2[i] = 0;
control = false;
}
else
{
matriz[i] = 0;
matriz2[i] = 1;
control = true;
}
}
entrada[ejemplos] = matriz;
entrada[ejemplos + 1] = matriz2;
entrada[ejemplos + 2] = matriz3;
//Creamos la matriz ideal
Double[][] ideal = new double[ejemplos + 3][];
for (int i = 0; i < ejemplos; i++)
{
ideal[i] = new double[] { 1 }
}
;
for (int i = ejemplos; i < ejemplos + 3; i++)
{
ideal[i] = new double[] { 0 }
}
;
//COMENZAMOS A CREAR LA RED NEURONAL!!!!
FeedforwardNetwork neurosis = new FeedforwardNetwork();
neurosis.AddLayer(new FeedforwardLayer(30));
neurosis.AddLayer(new FeedforwardLayer(30));
neurosis.AddLayer(new FeedforwardLayer(1));
neurosis.Reset();
Train entrenador = new HeatonResearchNeural.Feedforward.Train.Backpropagation.Backpropagation
(neurosis, entrada, ideal, 0.09, 0.9);
for (int e = 0; e < 1000; e++)
{
Console.WriteLine("Error: " + entrenador.Error);
entrenador.Iteration();
if (entrenador.Error < 0.012)
{
break;
}
}
redes[red - 1, pre] = neurosis;
}
private void trainNetworkBackprop()
{
Train train = new Backpropagation(this.network, this.input,
this.ideal, 0.001, 0.1);
int epoch = 1;
do
{
train.Iteration();
Console.WriteLine("Iteration #" + epoch + " Error:"
+ train.Error);
epoch++;
} while ((epoch < 5000) && (train.Error > 0.01));
}
private void trainNetworkBackprop()
{
Train train = new Backpropagation(this.network, this.input,
this.ideal, 0.000001, 0.1);
double lastError = Double.MaxValue;
int epoch = 1;
int lastAnneal = 0;
do
{
train.Iteration();
double error = train.Error;
Console.WriteLine("Iteration(Backprop) #" + epoch + " Error:"
+ error);
if (error > 0.05)
{
if ((lastAnneal > 100) && (error > lastError || Math.Abs(error - lastError) < 0.0001))
{
trainNetworkAnneal();
lastAnneal = 0;
}
}
lastError = train.Error;
epoch++;
lastAnneal++;
} while (train.Error > MAX_ERROR);
}
/// <summary>
/// Method that is called to start the incremental prune process.
/// </summary>
public void StartIncremental()
{
this.hiddenNeuronCount = 1;
this.cycles = 0;
this.done = false;
this.currentNetwork = new FeedforwardNetwork();
this.currentNetwork
.AddLayer(new FeedforwardLayer(this.train[0].Length));
this.currentNetwork.AddLayer(new FeedforwardLayer(
this.hiddenNeuronCount));
this.currentNetwork
.AddLayer(new FeedforwardLayer(this.ideal[0].Length));
this.currentNetwork.Reset();
this.backprop = new Backpropagation(this.currentNetwork, this.train,
this.ideal, this.rate, this.momentum);
}
/// <summary>
/// Internal method that is called at the end of each incremental cycle.
/// </summary>
protected void Increment()
{
bool doit = false;
if (this.markErrorRate == 0)
{
this.markErrorRate = this.error;
this.sinceMark = 0;
}
else
{
this.sinceMark++;
if (this.sinceMark > 10000)
{
if ((this.markErrorRate - this.error) < 0.01)
{
doit = true;
}
this.markErrorRate = this.error;
this.sinceMark = 0;
}
}
if (this.error < this.maxError)
{
this.done = true;
}
if (doit)
{
this.cycles = 0;
this.hiddenNeuronCount++;
this.currentNetwork = new FeedforwardNetwork();
this.currentNetwork.AddLayer(new FeedforwardLayer(
this.train[0].Length));
this.currentNetwork.AddLayer(new FeedforwardLayer(
this.hiddenNeuronCount));
this.currentNetwork.AddLayer(new FeedforwardLayer(
this.ideal[0].Length));
this.currentNetwork.Reset();
this.backprop = new Backpropagation(this.currentNetwork,
this.train, this.ideal, this.rate, this.momentum);
}
}
