package jsat.distributions.kernels;
import java.util.List;
import jsat.DataSet;
import jsat.distributions.Distribution;
import jsat.distributions.LogUniform;
import jsat.distributions.Uniform;
import jsat.linear.Vec;
import jsat.parameters.Parameterized;
/**
* The PUK kernel is an alternative to the RBF Kernel. By altering the
* {@link #setOmega(double) omega} parameter the behavior of the PUK kernel can
* be controlled. The {@link #setSigma(double) sigma} parameter works in the
* same way as the RBF Kernel.
*
* See: Üstün, B., Melssen, W. J.,&Buydens, L. M. C. (2006). Facilitating
* the application of Support Vector Regression by using a universal Pearson VII
* function based kernel. Chemometrics and Intelligent Laboratory Systems,
* 81(1), 29–40. doi:10.1016/j.chemolab.2005.09.003
*
* @author Edward Raff
*/
public class PukKernel extends BaseL2Kernel implements Parameterized
{
private static final long serialVersionUID = 8727097671803148320L;
private double sigma;
private double omega;
private double cnst;
/**
* Creates a new PUK Kernel
* @param sigma the width parameter of the kernel
* @param omega the shape parameter of the kernel
*/
public PukKernel(double sigma, double omega)
{
setSigma(sigma);
setOmega(omega);
}
/**
* Sets the omega parameter value, which controls the shape of the kernel
* @param omega the positive parameter value
*/
public void setOmega(double omega)
{
if(omega <= 0 || Double.isNaN(omega) || Double.isInfinite(omega))
throw new ArithmeticException("omega must be positive, not " + omega);
this.omega = omega;
this.cnst = Math.sqrt(Math.pow(2, 1/omega)-1);
}
public double getOmega()
{
return omega;
}
/**
* Sets the sigma parameter value, which controls the width of the kernel
* @param sigma the positive parameter value
*/
public void setSigma(double sigma)
{
if(sigma <= 0 || Double.isNaN(sigma) || Double.isInfinite(sigma))
throw new ArithmeticException("sigma must be positive, not " + sigma);
this.sigma = sigma;
}
public double getSigma()
{
return sigma;
}
@Override
public double eval(Vec a, Vec b)
{
return getVal(a.pNormDist(2.0, b));
}
@Override
public PukKernel clone()
{
return new PukKernel(sigma, omega);
}
@Override
public double eval(int a, Vec b, List qi, List extends Vec> vecs, List cache)
{
return getVal(Math.sqrt(getSqrdNorm(a, b, qi, vecs, cache)));
}
@Override
public double eval(int a, int b, List extends Vec> trainingSet, List cache)
{
return getVal(Math.sqrt(getSqrdNorm(b, b, trainingSet, cache)));
}
private double getVal(double pNormDist)
{
double tmp = 2*pNormDist*cnst/sigma;
return 1/Math.pow(1+tmp*tmp, omega);
}
/**
* Guesses the distribution to use for the ω parameter
*
* @param d the dataset to get the guess for
* @return the guess for the ω parameter
* @see #setOmega(double)
*/
public static Distribution guessOmega(DataSet d)
{
return new LogUniform(0.25, 50);
}
/**
* Guesses the distribution to use for the λ parameter
*
* @param d the dataset to get the guess for
* @return the guess for the λ parameter
* @see #setSigma(double)
*/
public static Distribution guessSigma(DataSet d)
{
return RBFKernel.guessSigma(d);
}
@Override
public boolean normalized()
{
return true;
}
}
Ads help maintain this website.