public void CreateBackPropagationTrainingAlgorithm(AnnBuild annComp, BackPropagation prop)
{
errorTarget = annComp.ErrorTarget;
network.InstantiateBackPropagationAlgorithm(
errorTarget, annComp.MaxEpochs, prop.LearningRate, prop.Beta);
network.BuildStatsListener();
}
public void TestMlpTrainingWthResilientPropPerformance()
{
using (wr = new StreamWriter(@"test.log", true, System.Text.Encoding.ASCII, 1024))
{
try
{
// Build network model
uint[] hlayers = new uint[1];
hlayers[0] = 4;
Build(3, 1, hlayers, 1);
wr.WriteLine("-> Test network performance for parity problem with rprop");
double targetError = 0.05;
uint epochs = 10000;
mlpNetwork.InstantiateResilientPropagationAlgorithm(targetError, epochs);
mlpNetwork.BuildStatsListener();
Task <bool> trainResult = mlpNetwork.TrainingAsync(parityInput, parityOutput, targetError);
DateTime start = DateTime.Now;
while (!trainResult.IsCompleted)
{
List <double[][]> output = mlpNetwork.GetOutputValues();
//wr.WriteLine("[" + (DateTime.Now-start).Milliseconds.ToString() + " ms] Output {0}", string.Join(";", output));
Thread.Sleep(2);
}
if (!trainResult.Result)
{
wr.WriteLine("-> Network training fail");
}
else
{
wr.WriteLine("-> Network trained! [Error: {0}, Epochs: {1}]", mlpNetwork.CurrentError, mlpNetwork.EpochsTraining);
wr.WriteLine("-> Validation:");
int i = 0;
foreach (double[] pattern in parityInputValidation)
{
double[] output = mlpNetwork.Exec(pattern);
wr.WriteLine("-> Input: " + pattern[0] + " " + pattern[1]);
wr.WriteLine("-> Output: " + output[0] + " (ideal: " + parityOutput[i++][0] + ")");
}
wr.WriteLine("-> Validation end");
}
} catch (Exception e)
{
wr.WriteLine(e.Message + Environment.NewLine + e.StackTrace);
}
}
}