package jsat.distributions;
import jsat.linear.Vec;
import static java.lang.Math.*;
import java.util.Random;
import static jsat.math.SpecialMath.*;
/**
*
* @author Edward Raff
*/
public class Gamma extends ContinuousDistribution
{
private static final long serialVersionUID = 6380493734491674483L;
private double k;
private double theta;
public Gamma(double k, double theta)
{
this.k = k;
this.theta = theta;
}
@Override
public double pdf(double x)
{
if(x < 0)
return 0;
return exp(logPdf(x));
}
@Override
public double logPdf(double x)
{
/*
* k - 1 / -x \
* x exp|-----|
* \theat/
* -----------------
* k
* Gamma(k) theta
*/
double p1 = -k *log(theta);
double p2 = k*log(x);
double p3 = -lnGamma(k);
double p4 = -x/theta;
double p5 = -log(x);
double pdf = p1+p2+p3+p4+p5;
if(Double.isNaN(pdf) || Double.isInfinite(pdf))//Bad extreme values when x is very small
return -Double.MAX_VALUE;
return pdf;
}
@Override
public double cdf(double x)
{
if(x < 0)
throw new ArithmeticException("CDF goes from 0 to Infinity, " + x + " is invalid");
return gammaP(k, x/theta);
}
@Override
public double invCdf(double p)
{
return invGammaP(p, k)*theta;
}
@Override
public double min()
{
return 0;
}
@Override
public double max()
{
return Double.POSITIVE_INFINITY;
}
@Override
public String getDistributionName()
{
return "Gamma";
}
@Override
public String[] getVariables()
{
return new String[] {"k", "theta"};
}
@Override
public double[] getCurrentVariableValues()
{
return new double[] {k, theta};
}
@Override
public void setVariable(String var, double value)
{
if(var.equals("k"))
k = value;
else if(var.equals("theta"))
theta = value;
}
@Override
public ContinuousDistribution clone()
{
return new Gamma(k, theta);
}
@Override
public void setUsingData(Vec data)
{
/*
* Using:
* mean = k*theat
* variance = k*theta^2
*
* k*theta^2 / (k*theta) = theta^2/theta = theta = mean/variance
*
*/
theta = data.variance()/data.mean();
k = data.mean()/theta;
}
@Override
public double mean()
{
return k * theta;
}
@Override
public double median()
{
return invGammaP(k, 0.5)*theta;
}
@Override
public double mode()
{
if(k < 1)
throw new ArithmeticException("No mode for k < 1");
return (k-1)*theta;
}
@Override
public double variance()
{
return k * theta*theta;
}
@Override
public double skewness()
{
return 2 / sqrt(k);
}
@Override
public double[] sample(int numSamples, Random rand)
{
/**
* See: Marsaglia, George, and Wai Wan Tsang. “A Simple Method for
* Generating Gamma Variables.” ACM Trans. Math. Softw. 26, no. 3
* (September 2000): 363–72. https://doi.org/10.1145/358407.358414.
*/
double[] toRet = new double[numSamples];
if (k >= 1.0)
{
double d = k - 1.0/3.0;
double c = 1.0/sqrt(9.0*d);
for(int i = 0; i < toRet.length; i++)
{
while(true)
{
double x = 0, xSqrd = 0;
double v = 0;
while (v <= 0.0)
{
x = rand.nextGaussian();
v = 1 + c*x;
}
v = v*v*v;
double u = rand.nextDouble();
xSqrd = x*x;
//Squeeze check done first to avoid expensieve logs
double squeezeCheck = 1.0 - 0.0331 * xSqrd * xSqrd;
if (u < squeezeCheck)
{
toRet[i] = theta*d*v;
break;
}//fail, now try logs if we must
else if( log(u) < 0.5 * xSqrd + d * (1.0 - v + log(v)))
{
toRet[i] = theta*d*v;
break;
}
}
}
}
else
{
Gamma shifted = new Gamma(k+1, 1.0);
double[] gs = shifted.sample(numSamples, rand);
for(int i = 0; i < toRet.length; i++)
toRet[i] = theta*gs[i]*pow(rand.nextDouble(), 1.0/k);
}
return toRet;
}
@Override
public int hashCode() {
final int prime = 31;
int result = 1;
long temp;
temp = Double.doubleToLongBits(k);
result = prime * result + (int) (temp ^ (temp >>> 32));
temp = Double.doubleToLongBits(theta);
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;
}
Gamma other = (Gamma) obj;
if (Double.doubleToLongBits(k) != Double.doubleToLongBits(other.k)) {
return false;
}
if (Double.doubleToLongBits(theta) != Double
.doubleToLongBits(other.theta)) {
return false;
}
return true;
}
}
Ads help maintain this website.