You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Alessandro Benedetti (JIRA)" <ji...@apache.org> on 2015/07/12 21:23:04 UTC
[jira] [Updated] (LUCENE-6654) KNearestNeighborClassifier not
taking in consideration Class ranking
[ https://issues.apache.org/jira/browse/LUCENE-6654?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Alessandro Benedetti updated LUCENE-6654:
-----------------------------------------
Attachment: LUCENE-6654.patch
- boost factor based on class single results scoring added
- 2 unit tests introduced to verify the proper behaviour
To try the effect, you can run the tests with original code, that will fail ignoring the class ranking
> KNearestNeighborClassifier not taking in consideration Class ranking
> --------------------------------------------------------------------
>
> Key: LUCENE-6654
> URL: https://issues.apache.org/jira/browse/LUCENE-6654
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/classification
> Affects Versions: 5.2.1
> Reporter: Alessandro Benedetti
> Priority: Minor
> Labels: classification, knn
> Attachments: LUCENE-6654.patch
>
>
> Currently the KNN Classifier assign the score for a ClassificationResult, based only on the frequency of the class in the top K results.
> This is conceptually a simplification.
> Actually the ranking must take a part.
> If not this can happen :
> Top 4
> 1) Class1
> 2) Class1
> 3) Class2
> 4) Class2
> As a result of this Top 4 , both the classes will have the same score.
> But the expected result is that Class1 has a better score, as the MLT score the documents accordingly.
--
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