Martin Wiesner created OPENNLP-1472:
---------------------------------------
Summary: Fix flaky (sub-) test of GradientDescentUtilsTest in
sandbox component 'nlp-utils'
Key: OPENNLP-1472
URL: https://issues.apache.org/jira/browse/OPENNLP-1472
Project: OpenNLP
Issue Type: Test
Affects Versions: 2.1.1
Reporter: Martin Wiesner
A flaky test has been discovered with recent modernizations of the sandbox
components. It lives in 'nlp-utils' and the problem signals a fragile test
setup or potentially numerical instabilities:
{quote}Error:
org.apache.opennlp.utils.regression.GradientDescentUtilsTest.testConvergence
Time elapsed: 1.156 s <<< ERROR!
[3270|https://github.com/apache/opennlp-sandbox/actions/runs/4177895676/jobs/7296072329#step:5:3271]java.lang.RuntimeException:
failed to converge at iteration 75129 with cost going from 5.010577258277822
to 5.010577258277823
[3271|https://github.com/apache/opennlp-sandbox/actions/runs/4177895676/jobs/7296072329#step:5:3272]
at
org.apache.opennlp.utils.regression.GradientDescentUtilsTest.testConvergence(GradientDescentUtilsTest.java:34)
{quote}
The problem (sometimes) occurs on different platforms / environments (Win, Lin,
Mac).
Aims:
* Investigate the cause of the flakyness
* Fix the test (environment / setup) so it runs reliably.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
