public void NotifyProgress_NotificationEpochs_is_zero()
{
Perceptron.interfaces.IBackPropagationAlgo trainer = new Perceptron.core.SGDTrainer();
Perceptron.entity.MultilayerPerceptron networkXOR = util.Helper.CreateXORNetwork();
trainer.Vectors = util.Helper.GenerateTrainingPointsForXor();
trainer.Perceptron = networkXOR;
trainer.LearningRate = 0.3;
trainer.MaxEpochs = 5000;
trainer.NotificationEpochs = 0;
int countOfEventsFired = 0;
Action <object, Perceptron.interfaces.NotifyProgressArgs> fnNotifyHandler = delegate(object sender, Perceptron.interfaces.NotifyProgressArgs args)
{
countOfEventsFired++;
};
trainer.OnNotifyProgressArgs += new EventHandler <Perceptron.interfaces.NotifyProgressArgs>(fnNotifyHandler);
trainer.Train();
var vectorsTest = util.Helper.GenerateTrainingPointsForXor();
Perceptron.entity.Vector[] vectorsWhichFailed = util.Helper.EvaluateVectors(trainer.Perceptron, vectorsTest);
Trace.WriteLine($"TEST DATA - The network generated {vectorsWhichFailed.Count()} errors after training.");
///
/// No events should have been fired
///
Assert.True(countOfEventsFired == 0);
Assert.True(vectorsWhichFailed.Count() == 0);
Assert.True(trainer.EpochsElapsed == trainer.MaxEpochs);
}
public void BasicXOR()
{
Perceptron.interfaces.IBackPropagationAlgo trainer = new Perceptron.core.SGDTrainer();
Perceptron.entity.MultilayerPerceptron networkXOR = util.Helper.CreateXORNetwork();
trainer.Vectors = util.Helper.GenerateTrainingPointsForXor();
trainer.Perceptron = networkXOR;
///
/// Randomize the network wts, ensure that the network produces atleast 1 error
/// so that we can test for success of training
///
while (true)
{
int errors = 0;
Perceptron.entity.Vector[] vectorsWhichFailed = util.Helper.EvaluateVectors(trainer.Perceptron, trainer.Vectors);
errors = vectorsWhichFailed.Count();
Trace.WriteLine($"TRAIN DATA - The network generated {errors} errors before training.");
if (errors > 0)
{
break;
}
Perceptron.core.Utils.RandomizeNetworkWeights(networkXOR);
}
string jsonUntrained = Perceptron.core.Utils.SaveNetworkToJson(networkXOR);
System.IO.File.WriteAllText("XOR.untrained.json", jsonUntrained);
trainer.NotificationEpochs = 10;
trainer.LearningRate = 0.3;
trainer.MaxEpochs = 5000;
trainer.Train();
///
/// Training complete - evaluate the results
///
{
int errors = 0;
Perceptron.entity.Vector[] vectorsWhichFailed = util.Helper.EvaluateVectors(trainer.Perceptron, trainer.Vectors);
errors = vectorsWhichFailed.Count();
Trace.WriteLine($"TRAIN DATA - The network generated {errors} errors after training.");
Assert.True(errors == 0);
}
///
/// Evaluate against test data
///
{
int errors = 0;
var vectorsTest = util.Helper.GenerateTrainingPointsForXor();
Perceptron.entity.Vector[] vectorsWhichFailed = util.Helper.EvaluateVectors(trainer.Perceptron, vectorsTest);
errors = vectorsWhichFailed.Count();
Trace.WriteLine($"TEST DATA - The network generated {errors} errors after training.");
Assert.True(errors == 0);
}
string jsonTrained = Perceptron.core.Utils.SaveNetworkToJson(networkXOR);
System.IO.File.WriteAllText("XOR.trained.json", jsonTrained);
}
public void NotifyProgress_StopTraining_WhenTrainingIsSuccessful()
{
Perceptron.interfaces.IBackPropagationAlgo trainer = new Perceptron.core.SGDTrainer();
Perceptron.entity.MultilayerPerceptron networkXOR = util.Helper.CreateXORNetwork();
trainer.Vectors = util.Helper.GenerateTrainingPointsForXor();
trainer.Perceptron = networkXOR;
trainer.LearningRate = 0.3;
trainer.MaxEpochs = 5000;
trainer.NotificationEpochs = 2;
int countOfEventsFired = 0;
///
/// Evaluate the success of the training data inside the event handler. When zero failures is reached halt the training
///
bool haltedInEventHandler = false;
Action <object, Perceptron.interfaces.NotifyProgressArgs> fnNotifyHandler = delegate(object sender, Perceptron.interfaces.NotifyProgressArgs args)
{
countOfEventsFired++;
double mseTotal = args.Vectors.Sum(v => v.MeanSquaredError);
int classificationerrors = 0;
foreach (var vec in args.Vectors)
{
double[] outputActual = vec.Outputs;
double[] outputActualQuantized = outputActual.Select(opt => (opt > 0.5) ? 1.0 : 0.0).ToArray();
double[] outputExpected = vec.Vector.Outputs;
if (outputActualQuantized.SequenceEqual(outputExpected) == false)
{
classificationerrors++;
}
}
Trace.WriteLine($" Epoch={trainer.EpochsElapsed} MSE={mseTotal}, Classification errors={classificationerrors}");
if (classificationerrors == 0)
{
Trace.WriteLine("Halting training because zero errors were reached");
args.Cancel = true;
haltedInEventHandler = true;
}
};
trainer.OnNotifyProgressArgs += new EventHandler <Perceptron.interfaces.NotifyProgressArgs>(fnNotifyHandler);
trainer.Train();
///
/// Some events should have been fired
///
Assert.True(haltedInEventHandler);
Assert.True(countOfEventsFired > 0);
Assert.True(countOfEventsFired == trainer.EpochsElapsed / 2);
}