[Weka-users] New releases: WEKA 3.8.7 and WEKA 3.9.7

[Eibe F. <eib...@wa...>]

Hi everyone,

New releases of WEKA 3.8 and 3.9 are now available. The most noteworthy additions to the collection of downloadables at

  https://waikato.github.io/weka-wiki/downloading_weka/

are platform‑specific distributions for ARM Windows and ARM Linux. As usual, we also provide an ARM version for macOS, along with Intel versions for all three platforms. In addition, there is a platform‑independent ZIP file that can be used with any Java 8 or later JVM on your system (although some WEKA packages beyond core WEKA may require recent Java versions).

All platform‑specific distributions are bundled with a corresponding Java 25 JRE provided by BellSoft. We have switched to BellSoft because they make available JREs for all six supported platforms that include JavaFX, which is required by some of the visualisation tools in WEKA packages.

The primary purpose of this release was to make WEKA work cleanly with Java 25 and to reduce the number of --add-opens arguments required when running WEKA from an operating system’s command‑line interface. We have also fixed a few other command‑line handling bugs and addressed some additional issues. Please see the changelog for a complete list of changes.

As with all recent versions of WEKA, the 3.8 (“stable”) and 3.9 (“development”) releases have identical functionality. We continue to maintain the 3.9 release series in anticipation of potential major future changes, which would be introduced there only.

WEKA has not been modularised and continues to run on the classpath rather than the module path. When running WEKA from an operating system’s command‑line interface without using the provided weka.sh (macOS, Linux) or RunWeka.bat (Windows) scripts included in the platform‑specific distributions, one --add-opens option is still required when using Java 25. For example, to run J48 from the command line:

  java -cp weka.jar --add-opens=java.base/java.lang=ALL-UNNAMED weka.Run J48 -t ...

Depending on whether code involving JavaFX or native library access is executed, the following additional flags may be used to suppress warnings:

  --enable-native-access=ALL-UNNAMED --enable-native-access=javafx.graphics

All WEKA packages should work on all supported platforms, including wekaPython and RPlugin. We have also made a netlibNative package for Windows ARM, and the netlibNative package for Linux now includes native code for both ARM and Intel. This enables fast native matrix algebra for schemes such as LinearRegression, GaussianProcesses, and related learners.

Cheers,
Eibe