private static void train(int ep)
{
// the input values
double[,] inputs =
{
{ 0, 0 },
{ 0, 1 },
{ 1, 0 },
{ 1, 1 }
};
// desired results
double[] results = { 0, 1, 1, 0 };
// creating the neurons
Neuron hiddenNeuron1 = new Neuron();
Neuron hiddenNeuron2 = new Neuron();
Neuron hiddenNeuron3 = new Neuron();
OutputNeuron outputNeuron = new OutputNeuron();
// random weights
hiddenNeuron1.randomizeWeights();
hiddenNeuron2.randomizeWeights();
hiddenNeuron3.randomizeWeights();
outputNeuron.randomizeWeights();
int epoch = 0;
Retry:
epoch++;
for (int i = 0; i < 4; i++) // very important, do NOT train for only one example
{
// 1) forward propagation (calculates output)
hiddenNeuron1.inputs = new double[] { inputs[i, 0], inputs[i, 1] };
hiddenNeuron2.inputs = new double[] { inputs[i, 0], inputs[i, 1] };
outputNeuron.inputs = new double[] { hiddenNeuron1.output, hiddenNeuron2.output };
Console.WriteLine("{0} xor {1} = {2}", inputs[i, 0], inputs[i, 1], outputNeuron.output());
Console.WriteLine(outputNeuron.output()[0]);
Console.WriteLine(outputNeuron.output()[1]);
Console.WriteLine(outputNeuron.output()[2]);
// 2) back propagation (adjusts weights)
// adjusts the weight of the output neuron, based on its error
outputNeuron.error = sigmoid.derivative(outputNeuron.output) * (results[i] - outputNeuron.output);
//get { return sigmoid.output(weights[0] * inputs[0] + weights[1] * inputs[1] + biasWeight); }
outputNeuron.adjustWeights();
// then adjusts the hidden neurons' weights, based on their errors
hiddenNeuron1.error = sigmoid.derivative(hiddenNeuron1.output) * outputNeuron.error * outputNeuron.weights[0];
hiddenNeuron2.error = sigmoid.derivative(hiddenNeuron2.output) * outputNeuron.error * outputNeuron.weights[1];
hiddenNeuron1.adjustWeights();
hiddenNeuron2.adjustWeights();
}
if (epoch < ep)
{
goto Retry;
}
OutPutNeuron1 = outputNeuron;
HiddenNeuron1 = hiddenNeuron1;
HiddenNeuron2 = hiddenNeuron2;
Console.ReadLine();
}
