package jsat.distributions.kernels;
import java.util.*;
import jsat.DataSet;
import jsat.distributions.Distribution;
import jsat.distributions.LogUniform;
import jsat.distributions.Uniform;
import jsat.linear.Vec;
import jsat.parameters.Parameter;
/**
* Provides an implementation of the Sigmoid (Hyperbolic Tangent) Kernel, which
* is of the form:
k(x, y) = tanh(alpha * < x, y > +c)
* Technically, this kernel is not positive definite.
*
* @author Edward Raff
*/
public class SigmoidKernel extends BaseKernelTrick
{
private static final long serialVersionUID = 8066799016611439349L;
private double alpha;
private double c;
/**
* Creates a new Sigmoid Kernel
* @param alpha the scaling factor for the dot product
* @param C the additive constant
*/
public SigmoidKernel(double alpha, double C)
{
this.alpha = alpha;
this.c = C;
}
/**
* Creates a new Sigmoid Kernel with a bias term of 1
* @param alpha the scaling factor for the dot product
*/
public SigmoidKernel(double alpha)
{
this(alpha, 1);
}
/**
* Sets the scaling factor for the dot product, this is equivalent to
* multiplying each value in the data set by a constant factor
* @param alpha the scaling factor
*/
public void setAlpha(double alpha)
{
if(Double.isInfinite(alpha) || Double.isNaN(alpha) || alpha == 0)
throw new IllegalArgumentException("alpha must be a real non zero value, not " + alpha);
this.alpha = alpha;
}
/**
* Returns the scaling parameter
* @return the scaling parameter
*/
public double getAlpha()
{
return alpha;
}
/**
* Sets the additive term, when set to one this is equivalent to adding a
* bias term of 1 to each vector. This is done after the scaling by
* {@link #setAlpha(double) alpha}.
* @param c the non negative additive term
*/
public void setC(double c)
{
if(c < 0 || Double.isNaN(c) || Double.isInfinite(c))
throw new IllegalArgumentException("C must be non negative, not " + c);
this.c = c;
}
/**
* Returns the additive constant
* @return the additive constant
*/
public double getC()
{
return c;
}
@Override
public double eval(Vec a, Vec b)
{
return Math.tanh(alpha*a.dot(b)+c);
}
/**
* Guesses a distribution for the α parameter
*
* @param d the data to get the guess for
* @return a distribution for the α parameter
*/
public static Distribution guessAlpha(DataSet d)
{
return new LogUniform(1e-12, 1e3);//from A Study on Sigmoid Kernels for SVM and the Training of non-PSD Kernels by SMO-type Methods
}
/**
* Guesses a distribution for the α parameter
*
* @param d the data to get the guess for
* @return a distribution for the α parameter
*/
public static Distribution guessC(DataSet d)
{
return new Uniform(-2.4, 2.4);//from A Study on Sigmoid Kernels for SVM and the Training of non-PSD Kernels by SMO-type Methods
}
@Override
public SigmoidKernel clone()
{
return new SigmoidKernel(alpha, c);
}
}
Ads help maintain this website.