// Simulate with test data, output: predicted class label
public int Simulate(Vector TestData, double Spread)
{
this.Spread = Spread;
for (int i = 0; i < NumberOfClass; i++)
{
int classLabel = this.ClassLabel[i];
for (int j = 0; j < GetNumberOfInstances(classLabel); j++)
{
ClassProbability[i] += Math.Exp(-Math.Pow((TrainingData[classLabel][j] - TestData).Norm(), 2) / Math.Pow((2 * Spread), 2));
}
ClassProbability[i] /= GetNumberOfInstances(classLabel);
}
double Max = ClassProbability[0];
int MaxIndex = 0;
for (int i = 1; i < NumberOfClass; i++)
{
if (ClassProbability[i] > Max)
{
MaxIndex = i;
Max = ClassProbability[i];
}
}
return ClassLabel[MaxIndex];
}
public int Simulate(Vector TestData)
{
double y = 0.0;
for (int i = 0; i < NumberOfCenter; i++)
{
double[] centerRow = CenterMatrix.GetRow(i);
Vector centerVector = new Vector(centerRow);
y += WeightMatrix.GetRow(i)[0] * GaussRBF(TestData, centerVector);
}
return (int)Math.Round(y);
}
public void AddData(Vector Data)
{
this.Data.Add(Data);
this.Dim = Data.Length;
}
public void Train()
{
// Vector to matrices
for (int i = 0; i < TrainingData.Count; i++)
{
for (int j = 0; j < TrainingData[i].Length; j++)
{
DataMatrix[i, j] = TrainingData[i][j];
}
TargetMatrix[i, 0] = TargetData[i];
}
// TODO: Center calculation using K-means
// Compute phi
PhiMatrix = new Matrix(NumberOfData, NumberOfCenter);
for (int i = 0; i < NumberOfData; i++)
{
for (int j = 0; j < NumberOfCenter; j++)
{
double[] dataRow = DataMatrix.GetRow(i);
double[] centerRow = CenterMatrix.GetRow(j);
Vector dataVector = new Vector(dataRow);
Vector centerVector = new Vector(centerRow);
PhiMatrix[i, j] = GaussRBF(dataVector, centerVector);
}
}
// Calculate weight
WeightMatrix = new Matrix(NumberOfCenter, 1);
WeightMatrix = (PhiMatrix.Transpose() * PhiMatrix).Invert() * PhiMatrix.Transpose() * TargetMatrix;
}
public double GaussRBF(Vector X, Vector Center)
{
return Math.Exp(-Math.Pow((X - Center).Norm(), 2) / Math.Pow((2 * this.Spread), 2));
}
public void AddTrainingData(int ClassLabel, Vector Data)
{
TargetData.Add(ClassLabel);
TrainingData.Add(Data);
}
public void AddTrainingData(int ClassLabel, Vector Data)
{
TrainingData[ClassLabel].Add(Data);
}
public double Dot(Vector V)
{
if (this.Length != V.Length) throw new Exception("Vector dimension mismatch!");
double sum = 0.0;
for (int i = 0; i < this.Length; i++)
{
sum += Value[i] * V[i];
}
return sum;
}
