[
https://issues.apache.org/jira/browse/LUCENE-8405?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16547491#comment-16547491
]
ASF subversion and git services commented on LUCENE-8405:
---------------------------------------------------------
Commit 331ccf391067b7775db371bec1ec8e8c9004e904 in lucene-solr's branch
refs/heads/master from [~jpountz]
[ https://git-wip-us.apache.org/repos/asf?p=lucene-solr.git;h=331ccf3 ]
LUCENE-8405: Remove TopDocs.maxScore.
> Remove TopHits.maxScore
> -----------------------
>
> Key: LUCENE-8405
> URL: https://issues.apache.org/jira/browse/LUCENE-8405
> Project: Lucene - Core
> Issue Type: Task
> Reporter: Adrien Grand
> Priority: Major
> Fix For: master (8.0)
>
> Attachments: LUCENE-8405.patch, LUCENE-8405.patch
>
>
> I would like to propose removing TopDocs.maxScore. The reasoning is that
> either you are sorting by score and then its value is easy to access via the
> score of the best hit. Or you sort by one or more fields and computing it is
> wasteful:
> - term frequencies and norms need to be read and decoded for every match
> - scores need to be computed on every match
> - early-termination optimizations are disabled
> It would be more efficient to collect hits twice: once with scores disabled
> to get the top hits, and once to get the best score which would run
> efficiently thanks to impacts and MAXSCORE, especially with a size of 1:
> {code:java}
> TopDocs topHits = searcher.search(query, 1);
> float maxScore = topHits.scoreDocs.length == 0 ? Float.NaN :
> topHits.scoreDocs[0].score;
> {code}
> The {{doDocScores}} option of TopFieldCollector has drawbacks as well but at
> least doesn't disable early-termination optimizations and doesn't require
> scores to be computed on every hit.
> As this would be a significant breaking change, I'm targeting 8.0.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
