public static Neuron MakeDummyNeuron(int inputSize)
{
Neuron dummy = new Neuron(LinearFunction.Instance,
FuzzySetIntersectOperation.Instance);
for (int i = 0; i < inputSize; i++)
dummy.Weights.Add(1.0);
return dummy;
}
public void TestInputOutput()
{
double[] input = { 1, 2, 3, 4, 5 };
double[] weights = { 1, 2, 3, 4, 5 };
double expected = 55;
Neuron neuron = new Neuron(LinearFunction.Instance);
for (int i = 0; i < weights.Length; i++)
neuron.Weights.Add(weights[i]);
neuron.Input(input);
Assert.AreEqual(expected, neuron.Output());
}
public void SetUp()
{
layer2 = new NeuronLayer();
layer1 = new NeuronLayer(layer2);
double[,] weights1 = new double[,] { { 1, 1,1 ,1 }, { 1, 1, 1, 1 },
{1, 1, 1, 1}, { 1, 1, 1, 1}};
double[,] weights2 = new double[,]{ { 1, 1, 1,1 }, { 1, 1, 1, 1 },
{1, 1, 1, 1}, { 1, 1, 1, 1}};
for (int i = 0; i < NEURON_SIZE; i++)
{
Neuron neuron1 = new Neuron(LinearFunction.Instance);
for(int j = 0; j <= weights1.GetUpperBound(0); j++)
neuron1.Weights.Add(weights1[i,j]);
layer1.AddNeuron(neuron1);
Neuron neuron2 = new Neuron(LinearFunction.Instance);
for (int j = 0; j <= weights2.GetUpperBound(0); j++)
neuron2.Weights.Add(weights2[i,j]);
layer2.AddNeuron(neuron2);
}
}
private void Initialize()
{
for (int i = 0; i < InputSize; ++i)
{
Neuron neuron = new Neuron(LinearFunction.Instance);
for (int j = 0; j < InputSize; ++j)
{
if (i == j)
neuron.Weights.Add(1.0);
else
neuron.Weights.Add(0.0);
}
f1.AddNeuron(neuron);
}
}
private void Process()
{
bool isSuccess = false;
while (!isSuccess)
{
// Select target weight
int winnerNeuronIndex = ArtA.WinningNeuronPos;
Neuron weightNeuron = null;
if (winnerNeuronIndex >= Map.Count)
{
// Create new weight neuron in mapfield.
weightNeuron = Map.Neurons(
Map.AddNeuron(FuzzyLogicUtil.MakeDummyNeuron(ArtB.InputSize)));
}
else
{
weightNeuron = Map.Neurons(winnerNeuronIndex);
}
if (training)
{
// Check vigilance condition
double[] artBValues = artB.Output;
weightNeuron.Input(artBValues);
double normOfSimilarity = weightNeuron.Output();
double testVigilance = normOfSimilarity / FuzzyLogicUtil.Norm(artBValues);
if (testVigilance >= Vigilance)
{
output = FuzzyLogicUtil.Intersect(weightNeuron.Weights.ToArray(), artBValues, ArtB.InputSize);
// Modify new mapfield weight.
for (int i = 0; i < weightNeuron.Weights.Count; ++i)
{
weightNeuron.Weights[i] = (Beta * output[i]) + ((1 - Beta) * weightNeuron.Weights[i]);
}
// Call ArtA to adjust its winner neuron weight.
bool isWeightAdjusted = false;
if (FastCommitedSlowLearningOption)
{
if (IsUnCommited(ArtA.F2.Neurons(ArtA.WinningNeuronPos)))
{
double beta = ArtA.Beta;
ArtA.Beta = 1.0;
artA.AdjustWeight();
ArtA.Beta = beta;
isWeightAdjusted = true;
}
}
if (!isWeightAdjusted)
{
ArtA.AdjustWeight();
}
// Call ArtB to adjust its winner neuron weight.
ArtB.AdjustWeight();
isSuccess = true;
}
else
{
double newVigilance = ArtA.WinningVigilance + ArtA.Choice;
ArtA.Vigilance = newVigilance > 1 ? 1.0 : newVigilance;
ArtA.Reset(winnerNeuronIndex);
}
}
else
{
output = weightNeuron.Weights.ToArray();
isSuccess = true;
}
}
training = false;
}
bool IsUnCommited(Neuron neuron)
{
bool isUnCommited = true;
if (neuron != null)
{
for (int i = 0; i < neuron.Weights.Count; i++)
{
if (neuron.Weights[i] != 1)
{
isUnCommited = false;
}
}
}
return isUnCommited;
}
public void RemoveNeuron(Neuron neuron)
{
InnerList.Remove(neuron);
}
public int AddNeuron(Neuron neuron)
{
return(InnerList.Add(neuron));
}