/* TEST NETWORK */
public double TestNetwork()
{
AverageError = 0;
Console.WriteLine("Start testing");
for (int i = 0; i < NumberOfExamples; i++)
{
//Console.WriteLine("Input number {0}", i + 1);
/* Add inputs to network and calculate outputs */
DigitNetwork.ResetAll();
DigitNetwork.SetInputValues(TestingData[i].ToArray <int>());
DigitNetwork.FeedResultsForward();
/*
* Console.WriteLine("Network result: {0}; expected result: {1}", DigitNetwork.GetResult(), ExpectedResults[i]);
* Console.WriteLine("Output neurons:");
* foreach (Neuron n in DigitNetwork.OutputLayer.Neurons)
* Console.Write("{0} ", n.GetOutput());
* Console.WriteLine();
*/
if (DigitNetwork.GetResult() == ExpectedResults[i])
{
CountCorrect++;
//Console.WriteLine("Correct identification, Correct = {0}", CountCorrect);
}
else
{
CountIncorrect++;
//Console.WriteLine("Wrong identification, Wrong = {0}", CountIncorrect);
}
AverageError += DigitNetwork.GetResult() - ExpectedResults[i];
}
AverageError /= NumberOfExamples;
Console.WriteLine("Average error: {0}", AverageError);
return(AverageError);
}
/* TRAIN NETWORK */
public double TrainNetwork(int numOfEpochs = 10)
{
double averageError, globalAverageError = 0;
int startingPoint;
Console.WriteLine("Start training");
for (int epoch = 0; epoch < numOfEpochs; epoch++)
{
averageError = 0;
startingPoint = new Random().Next(10);
for (int i = startingPoint; i < NumerOfExamples; i += 10)
{
//Console.WriteLine("Input number {0}", i+1);
/* Add inputs to network and calculate outputs */
DigitNetwork.ResetAll();
DigitNetwork.SetInputValues(TrainingData[i].ToArray <int>());
DigitNetwork.FeedResultsForward();
//Console.WriteLine("Network result: {0}; expected result: {1}", DigitNetwork.GetResult(), ExpectedResults[i]);
if (DigitNetwork.GetResult() == ExpectedResults[i])
{
CountCorrect++;
//Console.WriteLine("Correct identification, Correct = {0}", CountCorrect);
}
else
{
CountIncorrect++;
//Console.WriteLine("Wrong identification, Wrong = {0}", CountIncorrect);
}
averageError += DigitNetwork.GetResult() - ExpectedResults[i];
/* Calculate errors and propagate them back through the network, then adjust weights */
DigitNetwork.PropagateErrorsBack(LearningCoef, ExpectedResults[i]);
/*
* Console.WriteLine("Weights for the first hidden layer neurons:");
* foreach (Neuron neuron in ((SigmoidNeuronLayer)DigitNetwork.Layers[1]).Neurons)
* {
* Console.Write("{0}: ", ((SigmoidNeuron)neuron).ErrorCoef);
* foreach (Node node in neuron.Inputs.Keys)
* {
* Console.Write(neuron.Inputs[node]);
* Console.Write(" ");
* }
* Console.WriteLine();
* }
* Console.WriteLine("Weights for the second hidden layer neurons:");
* foreach (Neuron neuron in ((SigmoidNeuronLayer)DigitNetwork.Layers[2]).Neurons)
* {
* Console.Write("{0}: ", ((SigmoidNeuron)neuron).ErrorCoef);
* foreach (Node node in neuron.Inputs.Keys)
* {
* Console.Write(neuron.Inputs[node]);
* Console.Write(" ");
* }
* Console.WriteLine();
* }
* Console.WriteLine("Weights for the output neurons:");
* foreach (Neuron neuron in DigitNetwork.OutputLayer.Neurons)
* {
* Console.Write("{0}: ", ((SigmoidNeuron)neuron).ErrorCoef);
* foreach (Node node in neuron.Inputs.Keys)
* {
* Console.Write(neuron.Inputs[node]);
* Console.Write(" ");
* }
* Console.WriteLine();
* }
* */
}
averageError /= NumerOfExamples;
//Console.WriteLine("Average error: {0}", averageError);
globalAverageError += averageError;
}
globalAverageError /= numOfEpochs;
Console.WriteLine("Global average error: {0}", globalAverageError);
return(globalAverageError);
}
