[
https://issues.apache.org/jira/browse/LUCENE-1951?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12763118#action_12763118
]
Robert Muir commented on LUCENE-1951:
-------------------------------------
here are some stats for rewriting wildcards that should be prefix.
i query on a field with about ~10M numeric terms (a unique database id),
average length 10 characters or so.
i copied this into ramdirectory to try to rule out i/o a bit (its only 1GB
index and i use 4GB heap)
I look for all the ones starting with "1" (about 1.5 million of these). I did 3
runs, 100 queries each.
here are average times for each.
||Run||wildcardquery("1*")||prefixquery("1")
|1|1181ms|973ms
|2|1179ms|966ms
|3|1079ms|963ms
So, its not a big optimization, but seems consistent, and maybe more important
if avg term length is longer: in this case wildcard's comparison function might
have to do even more work.
I'll work on a patch to fix the boost/constant score and include a prefixquery
rewrite for this case.
> wildcardquery rewrite improvements
> ----------------------------------
>
> Key: LUCENE-1951
> URL: https://issues.apache.org/jira/browse/LUCENE-1951
> Project: Lucene - Java
> Issue Type: Improvement
> Components: Query/Scoring
> Reporter: Robert Muir
> Priority: Minor
>
> wildcardquery has logic to rewrite to termquery if there is no wildcard
> character, but
> * it needs to pass along the boost if it does this
> * if the user asked for a 'constant score' rewriteMethod, it should rewrite
> to a constant score query for consistency.
> additionally, if the query is really a prefixquery, it would be nice to
> rewrite to prefix query.
> both will enumerate the same number of terms, but prefixquery has a simpler
> comparison function.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
