You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "ASF subversion and git services (JIRA)" <ji...@apache.org> on 2013/11/22 09:30:39 UTC
[jira] [Commented] (LUCENE-4927) Prevent underflow in NB classifier
likelihood calculation
[ https://issues.apache.org/jira/browse/LUCENE-4927?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13829794#comment-13829794 ]
ASF subversion and git services commented on LUCENE-4927:
---------------------------------------------------------
Commit 1544433 from [~teofili] in branch 'dev/trunk'
[ https://svn.apache.org/r1544433 ]
LUCENE-4927 - switched to log prior/likelihood to avoid possible underflows
> Prevent underflow in NB classifier likelihood calculation
> ---------------------------------------------------------
>
> Key: LUCENE-4927
> URL: https://issues.apache.org/jira/browse/LUCENE-4927
> Project: Lucene - Core
> Issue Type: Bug
> Components: modules/classification
> Affects Versions: 4.2
> Reporter: Tommaso Teofili
> Assignee: Tommaso Teofili
> Fix For: 5.0
>
>
> Current likelihood calculation multiplies probabilities (whose values are between 0 and 1) thus having longish docs with unfrequent words for some class/category may lead to multiple _double_ multiplications to return 0 even if that's not the correct value (thus assigning such a class 0 probability too).
> Probably using loglikelihood and/or _BigDecimals_ may help.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org