Beast3

[walterxie]

Summary

Migrates nested-sampling from BEAST2 (2.7.x) / Ant to BEAST3 (2.8.0-beta7) / Maven. Tracked in #16. It now depends on model-selection:1.7.0-beta1.

Build system: Ant → Maven

• Replaced build.xml with pom.xml (Java 25 release target).
• Reorganized source tree into standard Maven layout (src/main/java, src/main/resources, src/test/java, src/test/resources).
• Added module-info.java (JPMS), requiring beast.base, model.selection, org.apache.commons.statistics.distribution, and org.apache.commons.math4.core, with optional BEAST FX/JavaFX for the GUI.
• Added .github/workflows/ci-publish.yml, src/assembly/beast-package.xml.
• Bumped version.xml to 1.3.0, depending on BEAST.base/BEAST.app >=2.8.0 and MODEL_SELECTION >=1.7.0.

Parameter → Tensor

Same BEAST3 type-system change as in model-selection. In NS.java, NIS.java, and NSOperatorSchedule.java, instanceof Parameter checks and Parameter.getDimension() calls became instanceof Tensor / Tensor.size() — StateNode dimension-counting logic is otherwise unchanged.

Commons Math → Commons Statistics / Commons Math 4

Two distinct replacements were needed, since BEAST3 dropped org.apache.commons.math entirely:

• NSLogAnalyser: BetaDistribution/BetaDistributionImpl → org.apache.commons.statistics.distribution.BetaDistribution (same pattern as model-selection).
• YuleModelNormalised: org.apache.commons.math3.util.FastMath → org.apache.commons.math4.core.jdkmath.AccurateMath — per the commons-math 4.0-beta1 changelog this is a straight s part of the legacy-module reorg, so behavior isunchanged; documented inline with a comment explaining why.

JUnit 4 → 5

All test classes use org.junit.jupiter.api.Test/Assertions; pom.xml adds junit-jupiter as a test dependency.

Test infrastructure duplicated to make the build compile

BEASTTestCase, XMLPathUtil, and ExampleXmlParsingTest (under src/test/java/test/beast/) were copied into this project's own test
sources rather than reused from beast-base, because s aren't published as a reusable test-jar artifact in the new modular Maven setup — duplicating them was the only way to get mvn test to compile.

YuleModelNormalised: moved main to a proper unit test

The original class had a public static void main(...) used only for manual smoke-testing, which pulled in a dependency on
BEASTTestCase inside the main sources. That method woved to a newYuleModelNormalisedTest.testYuleModelNormalised(). Since there was no previously-recorded expected value to migrate, the expected
logP (-10.018014963613476) was generated by running ST3 code itself — the test comment notes thisexplicitly ("computed from beast3 code on July 2026"), so it's a regression-guard baseline rather than an independently verified
value.

NS_4taxa_NormalBirthRate.xml: init issue, expected value diverges

Running the migrated XML under BEAST3 threw an initialization error from the RandomTree, fixed by adding rootHeight="0.19" to force a valid starting height. However, this changes the initial state enough that the nested-sampling evidence estimate Z no longer matches the original BEAST2 expected value:

• original expected: Z = -2349.5333124902536
• actual under BEAST3 (with the rootHeight fix): Z = -2359.7856198890186

Maven release beta1 is https://central.sonatype.com/artifact/io.github.beast2-dev/nested-sampling

---

TODO:

• Could not find class beast.base.core.BEASTInterface as service when loading BEAUti template
  Could not find class beast.base.core.BEASTInterface as service when loading BEAUti template #17