[MATH-1021] HypergeometricDistribution.sample suffers from integer overflow

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.3
• Assignee: Unassigned
• Reporter: Brian Bloniarz
• Created: 10/Aug/13 00:00
• Updated: 19/May/14 15:13
• Resolved: 29/Aug/13 21:35

Description

Hi, I have an application which broke when ported from commons math 2.2 to 3.2. It looks like the HypergeometricDistribution.sample() method doesn't work as well as it used to with large integer values – the example code below should return a sample between 0 and 50, but usually returns -50.

import org.apache.commons.math3.distribution.HypergeometricDistribution;

public class Foo {
  public static void main(String[] args) {
    HypergeometricDistribution a = new HypergeometricDistribution(
        43130568, 42976365, 50);
    System.out.printf("%d %d%n", a.getSupportLowerBound(), a.getSupportUpperBound()); // Prints "0 50"
    System.out.printf("%d%n",a.sample());                                             // Prints "-50"
  }
}

In the debugger, I traced it as far as an integer overflow in HypergeometricDistribution.getNumericalMean() – instead of doing

return (double) (getSampleSize() * getNumberOfSuccesses()) / (double) getPopulationSize();

it could do:

return getSampleSize() * ((double) getNumberOfSuccesses() / (double) getPopulationSize());

This seemed to fix it, based on a quick test.