[
https://issues.apache.org/jira/browse/LUCENE-8015?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Adrien Grand updated LUCENE-8015:
---------------------------------
Attachment: LUCENE-8015-test.patch
Given that this bug is not easy to reproduce due to how we finally cast double
scores to floats, which often returns the same value for consecutive values of
freq, I tried to hack the test framework to compare the produced doubles (see
attached patch - note this is for testing purpose only, I don't plan/want to
merge it). My assumption is that if we can reproduce the issue with doubles, it
means it can happen after a float cast as well, the scorer just needs to
produce a value that is close enough from the boundary between two floats so
that both values would round to different floats. And indeed tests fail
systematically with this patch. The bad news is that I can't think of a way to
fix the formula. Even if I put quite severe restrictions on the values that tfn
may take, there are still some special freq values that manage to prove the
score is not monotonic. Good news is that it doesn't make some other
SimilarityBase impls fail like the axiomatic ones.
> TestBasicModelIne.testRandomScoring failure
> -------------------------------------------
>
> Key: LUCENE-8015
> URL: https://issues.apache.org/jira/browse/LUCENE-8015
> Project: Lucene - Core
> Issue Type: Task
> Reporter: Adrien Grand
> Attachments: LUCENE-8015-test.patch, LUCENE-8015_test_fangs.patch
>
>
> reproduce with: ant test -Dtestcase=TestBasicModelIne
> -Dtests.method=testRandomScoring -Dtests.seed=86E85958B1183E93
> -Dtests.slow=true -Dtests.locale=vi-VN -Dtests.timezone=Pacific/Tongatapu
> -Dtests.asserts=true -Dtests.file.encoding=UTF8
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
