private static void Initialize(NeuralNet net, int randomSeed,
int inputNeuronCount, int hiddenNeuronCount, int outputNeuronCount)
{
#region Declarations
int i, j;
Random rand;
#endregion
#region Initialization
rand = new Random(randomSeed);
#endregion
#region Execution
net.m_inputLayer = new NeuralLayer();
net.m_outputLayer = new NeuralLayer();
net.m_hiddenLayer = new NeuralLayer();
for (i = 0; i < inputNeuronCount; i++)
{
net.m_inputLayer.Add(new Neuron(0));
}
for (i = 0; i < outputNeuronCount; i++)
{
net.m_outputLayer.Add(new Neuron(rand.NextDouble()));
}
for (i = 0; i < hiddenNeuronCount; i++)
{
net.m_hiddenLayer.Add(new Neuron(rand.NextDouble()));
}
// wire-up input layer to hidden layer
for (i = 0; i < net.m_hiddenLayer.Count; i++)
{
for (j = 0; j < net.m_inputLayer.Count; j++)
{
net.m_hiddenLayer[i].Input.Add(net.m_inputLayer[j], new NeuralFactor(rand.NextDouble()));
}
}
// wire-up output layer to hidden layer
for (i = 0; i < net.m_outputLayer.Count; i++)
{
for (j = 0; j < net.m_hiddenLayer.Count; j++)
{
net.m_outputLayer[i].Input.Add(net.HiddenLayer[j], new NeuralFactor(rand.NextDouble()));
}
}
#endregion
}
public static void CalculateAndAppendTransformation(NeuralNet net)
{
#region Declarations
int i, j;
INeuron outputNode, inputNode, hiddenNode;
#endregion
#region Execution
// adjust output layer weight change
for (j = 0; j < net.m_outputLayer.Count; j++)
{
outputNode = net.m_outputLayer[j];
for (i = 0; i < net.m_hiddenLayer.Count; i++)
{
hiddenNode = net.m_hiddenLayer[i];
outputNode.Input[hiddenNode].H_Vector += outputNode.Error * hiddenNode.Output;
}
outputNode.Bias.H_Vector += outputNode.Error * outputNode.Bias.Weight;
}
// adjust hidden layer weight change
for (j = 0; j < net.m_hiddenLayer.Count; j++)
{
hiddenNode = net.m_hiddenLayer[j];
for (i = 0; i < net.m_inputLayer.Count; i++)
{
inputNode = net.m_inputLayer[i];
hiddenNode.Input[inputNode].H_Vector += hiddenNode.Error * inputNode.Output;
}
hiddenNode.Bias.H_Vector += hiddenNode.Error * hiddenNode.Bias.Weight;
}
#endregion
}
private static void CalculateErrors(NeuralNet net, double[] desiredResults)
{
#region Declarations
int i, j;
double temp, error;
INeuron outputNode, hiddenNode;
#endregion
#region Execution
// Calcualte output error values
for (i = 0; i < net.m_outputLayer.Count; i++)
{
outputNode = net.m_outputLayer[i];
temp = outputNode.Output;
outputNode.Error = (desiredResults[i] - temp) * SigmoidDerivative(temp); //* temp * (1.0F - temp);
}
// calculate hidden layer error values
for (i = 0; i < net.m_hiddenLayer.Count; i++)
{
hiddenNode = net.m_hiddenLayer[i];
temp = hiddenNode.Output;
error = 0;
for (j = 0; j < net.m_outputLayer.Count; j++)
{
outputNode = net.m_outputLayer[j];
error += (outputNode.Error * outputNode.Input[hiddenNode].Weight) * SigmoidDerivative(temp);// *(1.0F - temp);
}
hiddenNode.Error = error;
}
#endregion
}
public static void PreparePerceptionLayerForPulse(NeuralNet net, double[] input)
{
#region Declarations
int i;
#endregion
#region Execution
if (input.Length != net.m_inputLayer.Count)
{
throw new ArgumentException(string.Format("Expecting {0} inputs for this net", net.m_inputLayer.Count));
}
// initialize data
for (i = 0; i < net.m_inputLayer.Count; i++)
{
net.m_inputLayer[i].Output = input[i];
}
#endregion
}
private void btnTrain_Click(object sender, EventArgs e)
{
#region Declarations
double ll, lh, hl, hh;
int count, iterations;
bool verbose;
double[][] input, output;
StringBuilder bld;
#endregion
#region Initialization
net = new NeuralNet();
bld = new StringBuilder();
input = new double[4][];
input[0] = new double[] { high, high };
input[1] = new double[] { low, high };
input[2] = new double[] { high, low };
input[3] = new double[] { low, low };
output = new double[4][];
output[0] = new double[] { low };
output[1] = new double[] { high };
output[2] = new double[] { high };
output[3] = new double[] { low };
verbose = false;
count = 0;
iterations = 5;
#endregion
#region Execution
// initialize with
// 2 perception neurons
// 2 hidden layer neurons
// 1 output neuron
net.Initialize(1, 2, 2, 1);
do
{
count++;
net.LearningRate = 3;
net.Train(input, output, TrainingType.BackPropogation, iterations);
net.PerceptionLayer[0].Output = low;
net.PerceptionLayer[1].Output = low;
net.Pulse();
ll = net.OutputLayer[0].Output;
net.PerceptionLayer[0].Output = high;
net.PerceptionLayer[1].Output = low;
net.Pulse();
hl = net.OutputLayer[0].Output;
net.PerceptionLayer[0].Output = low;
net.PerceptionLayer[1].Output = high;
net.Pulse();
lh = net.OutputLayer[0].Output;
net.PerceptionLayer[0].Output = high;
net.PerceptionLayer[1].Output = high;
net.Pulse();
hh = net.OutputLayer[0].Output;
#region verbose
if (verbose)
{
bld.Remove(0, bld.Length);
bld.Append("PERCEPTION LAYER <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<\n");
foreach (Neuron pn in net.PerceptionLayer)
{
AppendNeuronInfo(bld, pn);
}
bld.Append("\nHIDDEN LAYER <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<\n");
foreach (Neuron hn in net.HiddenLayer)
{
AppendNeuronInfo(bld, hn);
}
bld.Append("\nOUTPUT LAYER <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<\n");
foreach (Neuron on in net.OutputLayer)
{
AppendNeuronInfo(bld, on);
}
bld.Append("\n");
bld.Append("hh: \t").Append(hh.ToString()).Append(" \t< .5\n");
bld.Append("ll: \t").Append(ll.ToString()).Append(" \t< .5\n");
bld.Append("hl: \t").Append(hl.ToString()).Append(" \t> .5\n");
bld.Append("lh: \t").Append(lh.ToString()).Append(" \t> .5\n");
MessageBox.Show(bld.ToString());
}
#endregion
}
// really train this thing well...
while (hh > (mid + low) / 2 ||
lh < (mid + high) / 2 ||
hl < (mid + low) / 2 ||
ll > (mid + high) / 2);
net.PerceptionLayer[0].Output = low;
net.PerceptionLayer[1].Output = low;
net.Pulse();
ll = net.OutputLayer[0].Output;
net.PerceptionLayer[0].Output = high;
net.PerceptionLayer[1].Output = low;
net.Pulse();
hl = net.OutputLayer[0].Output;
net.PerceptionLayer[0].Output = low;
net.PerceptionLayer[1].Output = high;
net.Pulse();
lh = net.OutputLayer[0].Output;
net.PerceptionLayer[0].Output = high;
net.PerceptionLayer[1].Output = high;
net.Pulse();
hh = net.OutputLayer[0].Output;
bld.Remove(0, bld.Length);
bld.Append((count * iterations).ToString()).Append(" iterations required for training\n");
bld.Append("hh: ").Append(hh.ToString()).Append(" < .5\n");
bld.Append("ll: ").Append(ll.ToString()).Append(" < .5\n");
bld.Append("hl: ").Append(hl.ToString()).Append(" > .5\n");
bld.Append("lh: ").Append(lh.ToString()).Append(" > .5\n");
MessageBox.Show(bld.ToString());
#endregion
}
