[jira] [Comment Edited] (SPARK-7015) Multiclass to Binary Reduction

["Joseph K. Bradley (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-7015?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14504189#comment-14504189 ] 

Joseph K. Bradley edited comment on SPARK-7015 at 4/21/15 2:43 AM:
-------------------------------------------------------------------

+1  I'd strongly vote for supporting error-correcting output codes from early on.  It's not that much harder to implement, and it can perform much better in practice (and in theory).  I can provide some references if it'd be helpful.


was (Author: josephkb):
+1  I'd strongly vote for supporting error-correcting output codes from the beginning.  It's not that much harder to implement, and it can perform much better in practice (and in theory).  I can provide some references if it'd be helpful.

> Multiclass to Binary Reduction
> ------------------------------
>
>                 Key: SPARK-7015
>                 URL: https://issues.apache.org/jira/browse/SPARK-7015
>             Project: Spark
>          Issue Type: Improvement
>          Components: MLlib
>            Reporter: Ram Sriharsha
>            Assignee: Ram Sriharsha
>   Original Estimate: 336h
>  Remaining Estimate: 336h
>
> With the new Pipeline API, it is possible to seamlessly support machine learning reductions as meta algorithms.
> GBDT and SVM today are binary classifiers and we can implement multi class classification as a One vs All, or All vs All (or even more sophisticated reduction) using binary classifiers as primitives.
> This JIRA is to track the creation of a reduction API for multi class classification.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org