You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Robert Muir (JIRA)" <ji...@apache.org> on 2015/02/21 14:28:11 UTC

[jira] [Updated] (LUCENE-6274) SloppyPhrase approximations

     [ https://issues.apache.org/jira/browse/LUCENE-6274?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Robert Muir updated LUCENE-6274:
--------------------------------
    Attachment: LUCENE-6274.patch

Its difficult to go as far as we did for ExactPhrase initially here (use conjunction internally). 

An easier incremental step is to just do the "dead simple thing", to defer the positions matching. I think we should just do this here on this issue, and later try to do more intense refactoring (which might improve performance more, as the "conjunction" done in this scorer is strange).

I altered Adrien's task file from LUCENE-6244 to use phrases with slop of 5, (arbitrary, just to show it works) and ran on wikimedium10m:
{noformat}
                    Task   QPS trunk      StdDev   QPS patch      StdDev                Pct diff
   AndHighPhraseHighTerm        5.14      (5.0%)        5.14      (5.1%)    0.0% (  -9% -   10%)
    AndMedPhraseHighTerm       56.07      (3.4%)       63.38      (2.0%)   13.0% (   7% -   19%)
    AndHighPhraseMedTerm       21.30      (2.2%)       26.37      (1.8%)   23.8% (  19% -   28%)
     AndMedPhraseMedTerm       55.83      (3.0%)       70.08      (2.5%)   25.5% (  19% -   32%)
    AndHighPhraseLowTerm       45.19      (1.4%)       59.38      (1.1%)   31.4% (  28% -   34%)
     AndMedPhraseLowTerm      242.44      (2.7%)      332.56      (2.1%)   37.2% (  31% -   43%)
{noformat}

Still needs unit tests.

> SloppyPhrase approximations
> ---------------------------
>
>                 Key: LUCENE-6274
>                 URL: https://issues.apache.org/jira/browse/LUCENE-6274
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Robert Muir
>         Attachments: LUCENE-6274.patch, sloppy.tasks
>
>
> This scorer still does not support the approximations api to defer positions matching.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org