package jsat.distributions;
import jsat.linear.Vec;
import static java.lang.Math.*;
import static jsat.math.SpecialMath.*;
/**
*
* Also known as the F distribution.
*
* @author Edward Raff
*/
public class FisherSendor extends ContinuousDistribution
{
private static final long serialVersionUID = 7628304882101574242L;
double v1;
double v2;
public FisherSendor(double v1, double v2)
{
if(v1 <= 0)
throw new ArithmeticException("v1 must be > 0 not " + v1 );
if(v2 <= 0)
throw new ArithmeticException("v2 must be > 0 not " + v2 );
this.v1 = v1;
this.v2 = v2;
}
@Override
public double logPdf(double x)
{
if(x <= 0)
return 0;
double leftSide = v1/2 * log(v1) + v2/2*log(v2) - lnBeta(v1/2, v2/2);
double rightSide = (v1/2-1)*log(x) - (v1+v2)/2*log(v2+v1*x);
return leftSide+rightSide;
}
@Override
public double pdf(double x)
{
if(x <= 0)
return 0;
return exp(logPdf(x));
}
@Override
public double cdf(double x)
{
if(x <= 0)
return 0;
return betaIncReg(v1*x / (v1*x + v2), v1/2, v2/2);
}
@Override
public double invCdf(double p)
{
if(p < 0 || p > 1)
throw new ArithmeticException("Probability must be in the range [0,1], not" + p);
double u = invBetaIncReg(p, v1/2, v2/2);
return v2*u/(v1*(1-u));
}
@Override
public double min()
{
return 0;
}
@Override
public double max()
{
return Double.POSITIVE_INFINITY;
}
@Override
public String getDistributionName()
{
return "F";
}
@Override
public String[] getVariables()
{
return new String[]{"v1", "v2"};
}
@Override
public double[] getCurrentVariableValues()
{
return new double[]{v1, v2};
}
@Override
public void setVariable(String var, double value)
{
if (var.equals("v1"))
if (value > 0)
v1 = value;
else
throw new ArithmeticException("v1 must be > 0 not " + value);
else if (var.equals("v2"))
if (value > 0)
v2 = value;
else
throw new ArithmeticException("v2 must be > 0 not " + value );
}
@Override
public ContinuousDistribution clone()
{
return new FisherSendor(v1, v2);
}
@Override
public void setUsingData(Vec data)
{
double mu = data.mean();
//Only true if v2 > 2
double tmp = 2*mu / (-1 + mu);
if(tmp < 2)
{
return;//We couldnt approximate anything
}
else
{
v2 = tmp;
if(v2 < 4)
return;//We cant approximate v1
}
//only true if v2 > 4
double v2sqr = v2*v2;
double var = data.variance();
double denom = -2*v2sqr - 16*var + 20*v2*var - 8*v2sqr*var + v2sqr*v2*var;
v1 = 2*(-2*v2sqr + v2sqr*v2)/denom;
}
@Override
public double mean()
{
if(v2 <= 2)
return Double.NaN;
return v2/(v2-2);
}
@Override
public double median()
{
return (v2/v1)*(1.0/invBetaIncReg(0.5, v2/2, v1/2)-1);
}
@Override
public double mode()
{
if(v1 <= 2)
return Double.NaN;
return (v1-2)/v1*v2/(v2+2);
}
@Override
public double variance()
{
if(v2 <= 4)
return Double.NaN;
return 2 * v2*v2*(v1+v2-2) / (v1*pow(v2-2,2)*(v2-4));
}
@Override
public double skewness()
{
if(v2 <= 6)//Does not have a skewness for d2 <= 6
return Double.NaN;
double num = (2*v1+v2-2)*sqrt(8*(v2-4));
double denom = (v2-6)*sqrt(v1*(v1+v2-2));
return num/denom;
}
@Override
public int hashCode() {
final int prime = 31;
int result = 1;
long temp;
temp = Double.doubleToLongBits(v1);
result = prime * result + (int) (temp ^ (temp >>> 32));
temp = Double.doubleToLongBits(v2);
result = prime * result + (int) (temp ^ (temp >>> 32));
return result;
}
@Override
public boolean equals(Object obj) {
if (this == obj) {
return true;
}
if (obj == null) {
return false;
}
if (getClass() != obj.getClass()) {
return false;
}
FisherSendor other = (FisherSendor) obj;
if (Double.doubleToLongBits(v1) != Double.doubleToLongBits(other.v1)) {
return false;
}
if (Double.doubleToLongBits(v2) != Double.doubleToLongBits(other.v2)) {
return false;
}
return true;
}
}
Ads help maintain this website.