/// <summary>
/// <para>Returns an indication of similarity between vector <c>v1</c> and <c>v2</c>.</para>
/// <para>In this case, we use a sigmoid function.</para>
/// </summary>
protected override double KernelFunc(FeatureVector v1, FeatureVector v2)
{
double result = 0;
foreach (int f_i in v2.FeatureUnionWith(v2))
{
result += v1.Features[f_i] * v2.Features[f_i];
}
return(Math.Tanh((Gamma * result) + Coef));
}
/// <summary>
/// <para>Returns an indication of similarity between vector <c>v1</c> and <c>v2</c>.</para>
/// <para>In this case, we use a radial basis function.</para>
/// </summary>
protected override double KernelFunc(FeatureVector v1, FeatureVector v2)
{
double result = 0;
foreach (int f_i in v1.FeatureUnionWith(v2))
{
double diff = Math.Abs(v1.Features[f_i] - v2.Features[f_i]);
result += (diff * diff);
}
// Optimization:
// Here, you would usually square root the result to get the euclidean distance,
// but this kernel would ordinarily square the distance anyway.
// So, here we have just skipped these steps.
return(Math.Pow(Math.E, -1 * Gamma * result));
}
/// <summary>Calculates the Squared Euclidean Distance between two vectors.</summary>
private double CalculateAbsSquaredEuclideanDistance(FeatureVector v1, FeatureVector v2)
{
double distance = 0;
foreach (int f_i in v1.FeatureUnionWith(v2))
{
// Ignore OOV.
if (f_i >= NoOfFeatures)
{
continue;
}
double diff = (v2.Features[f_i] - v1.Features[f_i]);
distance += (diff * diff);
}
return(distance);
}
