You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@mahout.apache.org by "Ted Dunning (JIRA)" <ji...@apache.org> on 2011/02/01 19:22:28 UTC

[jira] Commented: (MAHOUT-525) Implement LatentFactorLogLinear models

    [ https://issues.apache.org/jira/browse/MAHOUT-525?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12989310#comment-12989310 ] 

Ted Dunning commented on MAHOUT-525:
------------------------------------

Robin,

I just saw your comment.

The reason is that this is an online learner.  Training data is available bit by bit, not all at once in these applications.

It would be reasonable to have a convenience method to train from the rows of a matrix chosen randomly, but I doubt that would be the wide usage.

> Implement LatentFactorLogLinear models
> --------------------------------------
>
>                 Key: MAHOUT-525
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-525
>             Project: Mahout
>          Issue Type: New Feature
>            Reporter: Ted Dunning
>             Fix For: 0.5
>
>
> The approach is based on the paper http://arxiv.org/abs/1006.2156
> For integration into Mahout, the idea would be to build an OnlineLearner that accept feature vectors where the first two elements are assumed to be the row
> and column id's.  The learning will proceed by stochastic gradient descent.

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