rcpeken commented on issue #16793:
URL: https://github.com/apache/lucene/issues/16793#issuecomment-6060246582
I looked into this. The two scores differ by exactly one float ulp: around
182.7 the ulp is about 1.53e-5, so the fixed 1e-5 delta effectively requires
exact equality, and a single rounding difference fails the test. Late
interaction scores are sums over query tokens, so they can get this large.
The seed also passes for me on main, and I couldn't pin down exactly where
the difference comes from. Rather than bumping to another fixed value like
#14954, I'd suggest a delta that scales with the score: `Math.max(1e-5f,
Math.abs(expected) * 1e-6f)`. If that sounds good I can open a PR against main;
since the failure came from branch_10x it would need a backport too.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
