[
https://issues.apache.org/jira/browse/LUCENE-8097?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Adrien Grand updated LUCENE-8097:
-
Attachment: LUCENE-8097.patch
Here is a patch. It uses the relative error bound documented in formula 3.5
from Higham, Nicholas J. (1993), "The accuracy of floating point summation",
SIAM Journal on Scientific Computing, which is {{E = (n-1) * u}} if all values
are positive, where n is the number of doubles that are summed up and u is the
unit roundoff. This implies that two sums cannot differ by more than 2*E.
> Implement Scorer.maxScore() on disjunctions
> ---
>
> Key: LUCENE-8097
> URL: https://issues.apache.org/jira/browse/LUCENE-8097
> Project: Lucene - Core
> Issue Type: Improvement
>Reporter: Adrien Grand
>Priority: Minor
> Fix For: master (8.0)
>
> Attachments: LUCENE-8097.patch
>
>
> We need to be careful with disjunctions since scores are not always summed up
> in the same order, which can result in different sums.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
-
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
