[
https://issues.apache.org/jira/browse/LUCENE-6926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15053794#comment-15053794
]
Paul Elschot commented on LUCENE-6926:
--------------------------------------
The patch uses only matchCost of req and excl, but it might be better to also
use their costs as a weight, just like a conjunction. The patch uses equal
weights for req and excl.
Here the conjunction is (req and (not excl)), and for weighting their costs are
req.cost() and (N - excl.cost()), with N as the number of docs in the segment.
This would boil down to normally using req first, as expected. Only when excl
matches a lot of docs, its weighting cost would be lower.
Would it be worthwhile to use such weights here?
See also LUCENE-6894 where the number of docs in a segment is used much more.
> Take matchCost into account for MUST_NOT clauses
> ------------------------------------------------
>
> Key: LUCENE-6926
> URL: https://issues.apache.org/jira/browse/LUCENE-6926
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-6926.patch
>
>
> ReqExclScorer potentially has two TwoPhaseIterators to check: the one for the
> positive clause and the one for the negative clause. It should leverage the
> match cost API to check the least costly one first.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
