private static void Main(string[] args)
{
var x = new XorTrainingData();
var w = new WeightLayer(3, 2);
var outputNode = new Node(3);
double error = 1.0;
int i = 0;
while (Math.Abs(error) > 0.005 && i++ < 70000)
{
error = Apply(x, w, outputNode);
if (i % 10000 == 0)
{
Console.WriteLine(string.Format("Iteration {0} {1}", i, error));
Check(x, w, outputNode);
Console.WriteLine();
}
}
Console.WriteLine(string.Format("Iterations = {0} {1}", i, error));
Check(x, w, outputNode);
Console.WriteLine("Press Enter to Exit.");
Console.ReadLine();
}
private static double Apply(XorTrainingData xorTrainingData, WeightLayer weightLayer, Node outputNode)
{
double totalError = 0;
foreach (var v in xorTrainingData.TrainingData)
{
// Forward
// for each hidden unit calculate its activation or value
for (int i = 0; i < weightLayer.Nodes.Length; i++)
{
weightLayer.Nodes[i].Calc(v.Inputs);
}
// Calculate the output or activation value for the network
outputNode.Calc(weightLayer.Values);
// Calculate the error (actual - target )
totalError += Math.Abs(v.Outputs[0] - outputNode.Value);
// Back Prop
// calculate error signal for output node
outputNode.ErrorSignal = (v.Outputs[0] - outputNode.Value) * outputNode.Value * (1 - outputNode.Value);
// calculate error term for hidden unit
for (int i = 0; i < weightLayer.Nodes.Length; i++)
{
weightLayer.Nodes[i].ErrorSignal = weightLayer.Nodes[i].Value * (1 - weightLayer.Nodes[i].Value) * outputNode.Weights[i] * outputNode.ErrorSignal;
}
// update the weights
// update the input to hidden layer weights
for (int i = 0; i < weightLayer.Nodes.Length; i++)
{
for (int j = 0; j < weightLayer.Nodes[i].Weights.Length; j++)
{
weightLayer.Nodes[i].Weights[j] += LearningRate * weightLayer.Nodes[i].ErrorSignal * v.Inputs[j];
}
}
// update the hidden layer to output weights
for (int i = 0; i < outputNode.Weights.Length; i++)
{
outputNode.Weights[i] += LearningRate * outputNode.ErrorSignal * weightLayer.Nodes[i].Value;
}
}
return(totalError);
}
private static void Check(XorTrainingData xorTrainingData, WeightLayer weightLayer, Node outputNode)
{
foreach (var v in xorTrainingData.TrainingData)
{
// for each hidden unit calculate its activation or value
for (int i = 0; i < weightLayer.Nodes.Length; i++)
{
weightLayer.Nodes[i].Calc(v.Inputs);
}
// Calculate the output or activation value for the network
outputNode.Calc(weightLayer.Values);
// Calculate the error (actual - target )
Console.WriteLine(string.Format("{0} {1} {2}", v.Outputs[0], outputNode.Value, (v.Outputs[0] - outputNode.Value)));
}
}