[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16572999#comment-16572999 ] ASF subversion and git services commented on LUCENE-8439: - Commit ba9b18f36743dc9674478dce5bbf2da509ef41c0 in lucene-solr's branch refs/heads/master from [~jim.ferenczi] [ https://git-wip-us.apache.org/repos/asf?p=lucene-solr.git;h=ba9b18f ] LUCENE-8439: Disjunction max queries can skip blocks to select the top documents when the total hit count is not required > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch, LUCENE-8439.patch, LUCENE-8439.patch, > LUCENE-8439.patch, LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16572938#comment-16572938 ] Adrien Grand commented on LUCENE-8439: -- FYI the patch includes some unintended changes in the solr/ directory, but changes under lucene/ look good to me! > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch, LUCENE-8439.patch, LUCENE-8439.patch, > LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16572934#comment-16572934 ] Adrien Grand commented on LUCENE-8439: -- +1 > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch, LUCENE-8439.patch, LUCENE-8439.patch, > LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16572889#comment-16572889 ] Jim Ferenczi commented on LUCENE-8439: -- I need to access the BlockMaxDISI in the DisjunctionMaxScorer but it can work with a simple getter in the DisjunctionScorer. I pushed a new patch that replaces the wrapTopScoresApprox with this approach. > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch, LUCENE-8439.patch, LUCENE-8439.patch, > LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16571970#comment-16571970 ] Adrien Grand commented on LUCENE-8439: -- I'm curious why you introduced {{wrapTopScoresApprox}} instead of directly wrapping the approximation with a BlockMaxDISI in {{DisjunctionScorer}}? > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch, LUCENE-8439.patch, LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16565438#comment-16565438 ] Jim Ferenczi commented on LUCENE-8439: -- Thanks for looking Adrien. I pushed a new patch that factors out a BlockMaxDISI implementation and uses it only if the score mode is set to TOP_SCORES. {quote} I guess it works well because scores on title dominate the overall score? {quote} Yes, this optim works best when the small field (here title) dominates the overall score. > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch, LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16564252#comment-16564252 ] Adrien Grand commented on LUCENE-8439: -- Pretty cool. I guess it works well because scores on title dominate the overall score? Should we factor out a DocIdSetIterator implementation similar to ImpactsDISI but that would take a Scorer rather than an ImpactsSource to reuse in ReqOptSumScorer, DisjunctionMaxScorer and probably other scorers (ConstantScoreScorer and CoveringScorer come to mind). Maybe we should also replace {{boolean needsScores}} with {{ScoreMode scoreMode}} so that we would only wrap the approximation when top scores are needed rather than all scores? > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8439) DisjunctionMaxScorer should leverage sub scorers' per-block max scores
[ https://issues.apache.org/jira/browse/LUCENE-8439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16563761#comment-16563761 ] Jim Ferenczi commented on LUCENE-8439: -- Here is a patch that uses the same strategy to find blocks than LUCENE-8204. I ran a benchmark using wikibigall and disjunction max queries on the Low/Med/High terms over the body and title fields. Results below show a nice speed up with any tiebreaker value (except 1 since we rewrite to a boolean query in this case): {noformat} TaskQPS lucene_baseline StdDevQPS lucene_candidate StdDev Pct diff LowTerm 121.15 (3.3%) 838.54 (33.6%) 592.1% ( 537% - 650%) MedTerm 25.52 (5.1%) 385.07(114.0%) 1409.1% (1227% - 1610%) HighTerm9.62 (5.6%) 183.10(162.7%) 1803.6% (1548% - 2089%) {noformat} > DisjunctionMaxScorer should leverage sub scorers' per-block max scores > -- > > Key: LUCENE-8439 > URL: https://issues.apache.org/jira/browse/LUCENE-8439 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Jim Ferenczi >Priority: Minor > Attachments: LUCENE-8439.patch > > > This issue is similar to https://issues.apache.org/jira/browse/LUCENE-8204 > but for the DisjunctionMaxScorer. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org