[jira] [Updated] (MAHOUT-678) NullPointerException while using MixedGradient with SGD algorithm

["Ted Dunning (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/MAHOUT-678?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Ted Dunning updated MAHOUT-678:
-------------------------------

    Fix Version/s: 0.6

This needs fixing on at least two levels.  First, the documentation should describe why only the binomial case makes sense for ranking and secondly, the code should detect mis-use and complain.

> NullPointerException while using MixedGradient with SGD algorithm
> -----------------------------------------------------------------
>
>                 Key: MAHOUT-678
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-678
>             Project: Mahout
>          Issue Type: Bug
>          Components: Classification
>    Affects Versions: 0.5
>            Reporter: Stanley Xu
>             Fix For: 0.6
>
>         Attachments: nullpointerexceptionbugfix.diff
>
>   Original Estimate: 2h
>  Remaining Estimate: 2h
>
> I am trying to use the MixedGradient in OnlineLogisticRegression algorithm. But I will get an NullPointerException randomly if I set the alpha larger than 0. 
> I checked the code and found that in the RankingGradient used by MixedGradient, it assume that the target category should be only 2, rather than multiple. And the rank gradient should only be used once the Gradient object knew both the positive and negative targets. I created a simple patch to make it workable, but I am not really understand the MixedGradient method deeply, please check the patch carefully to see if it really works correctly.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira