This is an automated email from the ASF dual-hosted git repository.
aherbert pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/commons-statistics.git
commit 3cb1c10953fadf74c4d86582601167e51d1afae0
Author: aherbert <aherb...@apache.org>
AuthorDate: Tue Feb 21 12:59:21 2023 +0000
Updated many ties test to test an estimated p-value
---
.../inference/KolmogorovSmirnovTestTest.java | 36 +++++++++++-----------
1 file changed, 18 insertions(+), 18 deletions(-)
diff --git
a/commons-statistics-inference/src/test/java/org/apache/commons/statistics/inference/KolmogorovSmirnovTestTest.java
b/commons-statistics-inference/src/test/java/org/apache/commons/statistics/inference/KolmogorovSmirnovTestTest.java
index 168c17c..9fa95c9 100644
---
a/commons-statistics-inference/src/test/java/org/apache/commons/statistics/inference/KolmogorovSmirnovTestTest.java
+++
b/commons-statistics-inference/src/test/java/org/apache/commons/statistics/inference/KolmogorovSmirnovTestTest.java
@@ -1096,24 +1096,24 @@ class KolmogorovSmirnovTestTest {
Assertions.assertTrue(r.getUpperD() > r.getStatistic(), "upper D");
Assertions.assertTrue(r.getUpperPValue() < r.getPValue(), "upper D");
- // XXX extract to a test for estimateP
-// // The p-value is not close to p(D=0.0640) due to random tie
resolution.
-// // exactP = 0.9659836534406034 (with dnm = (long) (d * n * m))
-// // estimateP = 0.558364 (with 1000000 iterations)
-// // Here we repeat call the test using different random seeds
-// // for tie resolution and take an average p-value.
-// // The relative error is stable around 0.065. Examples with a
randomly seeded RNG:
-// // Trials Relative error
-// // 1000 ~0.0622
-// // 10000 ~0.0651
-// // 100000 ~0.0628
-// // For robustness use a fixed seed with a small number of trials
-//
-// final double p = RandomSource.SPLIT_MIX_64.create(12345)
-// .longs(100)
-// .mapToDouble(seed ->
KolmogorovSmirnovTest.kolmogorovSmirnovTest(x, y, false, seed))
-// .average().getAsDouble();
-// TestUtils.assertProbability(0.558364, p, 0.065, "p-value");
+ // Test a p-value estimation:
+ // o.a.c.m.stat.inference.KolmogorovSmirnovTest.bootstrap = 0.558364
(with 1000000 iterations)
+ //
+ // R 3.4.0, ks.boot implemented in Matching (Version 4.10.8).
+ // require('Matching')
+ // ks.boot(x, y, nboots=1000000)
+ // ks.boot.pvalue: 0.55919
+ //
+ // The p-value is very variable so relative epsilon is for 2
significant digits.
+ // Even with such low tolerance the test is sensitive to the RNG and a
fixed seed
+ // RNG is used for each case. This method is a general estimate of the
p-value.
+ final TwoResult r2 = KolmogorovSmirnovTest.withDefaults()
+ .with(PValueMethod.ESTIMATE)
+ .with(RandomSource.SPLIT_MIX_64.create(12345))
+ .test(x, y);
+ Assertions.assertEquals(r.getStatistic(), r2.getStatistic(),
"statistic mismatch");
+ TestUtils.assertRelativelyEquals(0.558364, r2.getPValue(), 1e-2,
"estimated p-value");
+ TestUtils.assertRelativelyEquals(0.55919, r2.getPValue(), 1e-2,
"estimated p-value");
}
@Test
