[MATH-1197] 2-sample KS statistic was wrong in case of ties.
This commit is contained in:
Thomas Neidhart
parent
9a0d061981
commit
870e1d3d98
|
|
@ -50,7 +50,11 @@ If the output is not quite correct, check for invisible trailing spaces!
|
|||
<title>Commons Math Release Notes</title>
|
||||
</properties>
|
||||
<body>
|
||||
<release version="TBD" date="TBD" description="TBD">
|
||||
<release version="3.6" date="XXXX-XX-XX" description="">
|
||||
<action dev="tn" type="fix" issue="MATH-1197">
|
||||
Computation of 2-sample Kolmogoriv-Smirnov statistic in case of ties
|
||||
was not correct.
|
||||
</action>
|
||||
</release>
|
||||
<release version="3.5" date="2015-04-17" description="
|
||||
This is a minor release: It combines bug fixes and new features.
|
||||
|
|
|
|||
|
|
@ -298,9 +298,13 @@ public class KolmogorovSmirnovTest {
|
|||
double supD = 0d;
|
||||
// First walk x points
|
||||
for (int i = 0; i < n; i++) {
|
||||
final double cdf_x = (i + 1d) / n;
|
||||
final int yIndex = Arrays.binarySearch(sy, sx[i]);
|
||||
final double cdf_y = yIndex >= 0 ? (yIndex + 1d) / m : (-yIndex - 1d) / m;
|
||||
final double x_i = sx[i];
|
||||
// ties can be safely ignored
|
||||
if (i > 0 && x_i == sx[i-1]) {
|
||||
continue;
|
||||
}
|
||||
final double cdf_x = edf(x_i, sx);
|
||||
final double cdf_y = edf(x_i, sy);
|
||||
final double curD = FastMath.abs(cdf_x - cdf_y);
|
||||
if (curD > supD) {
|
||||
supD = curD;
|
||||
|
|
@ -308,9 +312,13 @@ public class KolmogorovSmirnovTest {
|
|||
}
|
||||
// Now look at y
|
||||
for (int i = 0; i < m; i++) {
|
||||
final double cdf_y = (i + 1d) / m;
|
||||
final int xIndex = Arrays.binarySearch(sx, sy[i]);
|
||||
final double cdf_x = xIndex >= 0 ? (xIndex + 1d) / n : (-xIndex - 1d) / n;
|
||||
final double y_i = sy[i];
|
||||
// ties can be safely ignored
|
||||
if (i > 0 && y_i == sy[i-1]) {
|
||||
continue;
|
||||
}
|
||||
final double cdf_x = edf(y_i, sx);
|
||||
final double cdf_y = edf(y_i, sy);
|
||||
final double curD = FastMath.abs(cdf_x - cdf_y);
|
||||
if (curD > supD) {
|
||||
supD = curD;
|
||||
|
|
@ -319,6 +327,24 @@ public class KolmogorovSmirnovTest {
|
|||
return supD;
|
||||
}
|
||||
|
||||
/**
|
||||
* Computes the empirical distribution function.
|
||||
*
|
||||
* @param x the given x
|
||||
* @param samples the observations
|
||||
* @return the empirical distribution function \(F_n(x)\)
|
||||
*/
|
||||
private double edf(final double x, final double[] samples) {
|
||||
final int n = samples.length;
|
||||
int index = Arrays.binarySearch(samples, x);
|
||||
if (index >= 0) {
|
||||
while(index < (n - 1) && samples[index+1] == x) {
|
||||
++index;
|
||||
}
|
||||
}
|
||||
return index >= 0 ? (index + 1d) / n : (-index - 1d) / n;
|
||||
}
|
||||
|
||||
/**
|
||||
* Computes the <i>p-value</i>, or <i>observed significance level</i>, of a one-sample <a
|
||||
* href="http://en.wikipedia.org/wiki/Kolmogorov-Smirnov_test"> Kolmogorov-Smirnov test</a>
|
||||
|
|
@ -429,7 +455,7 @@ public class KolmogorovSmirnovTest {
|
|||
return 1;
|
||||
}
|
||||
if (exact) {
|
||||
return exactK(d,n);
|
||||
return exactK(d, n);
|
||||
}
|
||||
if (n <= 140) {
|
||||
return roundedK(d, n);
|
||||
|
|
@ -834,8 +860,13 @@ public class KolmogorovSmirnovTest {
|
|||
* @throws TooManyIterationsException if the series does not converge
|
||||
*/
|
||||
public double ksSum(double t, double tolerance, int maxIterations) {
|
||||
if (t == 0.0) {
|
||||
return 1.0;
|
||||
}
|
||||
|
||||
// TODO: for small t (say less than 1), the alternative expansion in part 3 of [1]
|
||||
// from class javadoc should be used.
|
||||
|
||||
final double x = -2 * t * t;
|
||||
int sign = -1;
|
||||
long i = 1;
|
||||
|
|
|
|||
|
|
@ -302,6 +302,69 @@ public class KolmogorovSmirnovTestTest {
|
|||
}
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testTwoSampleWithManyTies() {
|
||||
// MATH-1197
|
||||
final double[] x = {
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 2.202653, 2.202653, 2.202653, 2.202653, 2.202653,
|
||||
2.202653, 2.202653, 2.202653, 2.202653, 2.202653, 2.202653,
|
||||
2.202653, 2.202653, 2.202653, 2.202653, 2.202653, 2.202653,
|
||||
2.202653, 2.202653, 2.202653, 2.202653, 2.202653, 2.202653,
|
||||
2.202653, 2.202653, 2.202653, 2.202653, 2.202653, 2.202653,
|
||||
2.202653, 2.202653, 2.202653, 2.202653, 2.202653, 2.202653,
|
||||
3.181199, 3.181199, 3.181199, 3.181199, 3.181199, 3.181199,
|
||||
3.723539, 3.723539, 3.723539, 3.723539, 4.383482, 4.383482,
|
||||
4.383482, 4.383482, 5.320671, 5.320671, 5.320671, 5.717284,
|
||||
6.964001, 7.352165, 8.710510, 8.710510, 8.710510, 8.710510,
|
||||
8.710510, 8.710510, 9.539004, 9.539004, 10.720619, 17.726077,
|
||||
17.726077, 17.726077, 17.726077, 22.053875, 23.799144, 27.355308,
|
||||
30.584960, 30.584960, 30.584960, 30.584960, 30.751808
|
||||
};
|
||||
|
||||
final double[] y = {
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,
|
||||
0.000000, 0.000000, 0.000000, 2.202653, 2.202653, 2.202653,
|
||||
2.202653, 2.202653, 2.202653, 2.202653, 2.202653, 3.061758,
|
||||
3.723539, 5.628420, 5.628420, 5.628420, 5.628420, 5.628420,
|
||||
6.916982, 6.916982, 6.916982, 10.178538, 10.178538, 10.178538,
|
||||
10.178538, 10.178538
|
||||
};
|
||||
|
||||
final KolmogorovSmirnovTest test = new KolmogorovSmirnovTest();
|
||||
|
||||
Assert.assertEquals(0.0640394088, test.kolmogorovSmirnovStatistic(x, y), 1e-6);
|
||||
Assert.assertEquals(0.9792777290, test.kolmogorovSmirnovTest(x, y), 1e-6);
|
||||
|
||||
}
|
||||
|
||||
/**
|
||||
* Verifies the inequality exactP(criticalValue, n, m, true) < alpha < exactP(criticalValue, n,
|
||||
* m, false).
|
||||
|
|
|
|||
Loading…
Reference in New Issue