MLPTest and CANDTest fail on GitHub job runners. #322
Description
Running the following:
mvn test
With the following environment (ubuntu-latest):
Apache Maven 3.9.11 (3e54c93a704957b63ee3494413a2b544fd3d825b)
Maven home: /usr/share/apache-maven-3.9.11
Java version: 21.0.8, vendor: Azul Systems, Inc., runtime: /opt/hostedtoolcache/Java_Zulu_jdk/21.0.8-9/x64
Default locale: en, platform encoding: UTF-8
OS name: "linux", version: "6.14.0-1012-azure", arch: "amd64", family: "unix"
Results in this errors:
testRegression(moa.classifiers.deeplearning.MLPTest) Time elapsed: 27.61 sec <<< FAILURE!
junit.framework.AssertionFailedError: Output differs:
7,8c7,8
< 0: -0.07205777
< 1: 0.07205255
---
> 0: -0.07205774
> 1: 0.07205258
21,22c21,22
< 0: -0.07205777
< 1: 0.07205255
---
> 0: -0.07205774
> 1: 0.07205258
35,36c35,36
< 0: -0.07205777
< 1: 0.07205255
---
> 0: -0.07205774
...
testRegression(moa.classifiers.deeplearning.CANDTest) Time elapsed: 87.771 sec <<< FAILURE!
junit.framework.AssertionFailedError: Output differs:
7,8c7,8
< 0: 0.0023892
< 1: -0.0023916
---
> 0: -0.30183133
> 1: 0.30182815
21,22c21,22
< 0: 0.0023892
< 1: -0.0023916
---
> 0: -0.30183133
> 1: 0.30182815
35,36c35,36
< 0: -0.40148267
< 1: 0.4014819
---
> 0: -0.30183133
> 1: 0.30182815
...
Things run correctly locally and the problem persists for different java distributions. It is potentially hardware related if the deep learning code uses BLAS?