package jsat.distributions.empirical.kernelfunc;

import java.io.Serializable;

/**
 * Class for representing one dimensional kernel functions. Since they require 
 * no parameters and have no need for duplication, its is advised to make 
 * them singletons. 
 * 
 * See http://en.wikipedia.org/wiki/Kernel_(statistics)
 * 
 * @author Edward Raff
 */
public interface KernelFunction extends Serializable
{
    /**
     * Returns the weight to be applied to a sample for the normalized distance of two data points. 
     * @param u the distance of the data points
     * @return the value in [0, 1) of the amount of weight to give to the sample based on its distance 
     */
    public double k(double u);
    /**
     * Computes the value of the finite integral from -Infinity up to the value u, of the function given by {@link #k(double) }
     * @param u the distance of the data points
     * @return the value of the integration
     */
    public double intK(double u);
    
    /**
     * 
     * Returns the value of the derivative at a point, k'(u)
     * @param u the distance of the data points
     * @return the value of the derivative at u
     */
    public double kPrime(double u);
    
    /**
     * Returns the variance of the kernel function
     * @return the variance of the kernel function
     */
    public double k2();
    
    /**
     * As the value of |u| for the kernel function approaches infinity, the
     * value of k(u) approaches zero. This function returns the minimal 
     * absolute value of u for which k(u) returns 0
     * 
     * @return the first value for which k(u) = 0
     */
    public double cutOff();
    
    /**
     * Returns the name of this kernel function
     * @return the name of this kernel function 
     */
    @Override
    public String toString();
}

 

Ads help maintain this website.