You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mahout.apache.org by "Dmitriy Lyubimov (JIRA)" <ji...@apache.org> on 2010/11/14 08:56:14 UTC

[jira] Updated: (MAHOUT-376) Implement Map-reduce version of stochastic SVD

     [ https://issues.apache.org/jira/browse/MAHOUT-376?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dmitriy Lyubimov updated MAHOUT-376:
------------------------------------

    Attachment: QR decomposition for Map.pdf

I am currently working to drive the working prototype the version with standalone thin QR step which would be memory independent of num of rows in A and would result into BBt eigensolution of (k+p)x(k+p) dimensionality only, the rest being driven by Map Reduce.  I've got single stream version working seemingly well, here's the update on this WIP. Although in the end i am not sure it would offer any real-life improvement, as it would seem to require a second pass over A as it is not possible to finish Q^t x B computation in single step with this approach. 

Still, after 2 passes, we should be done with eigen values (and perhaps (k+p)x(k+p) dimensionality for eigensolver input would allow us to increase oversampling p somewhat, hence precision). Hard to see from here yet though. Additional (optional) MR steps would only be needed if U or V is or both are desired.

> Implement Map-reduce version of stochastic SVD
> ----------------------------------------------
>
>                 Key: MAHOUT-376
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-376
>             Project: Mahout
>          Issue Type: Improvement
>          Components: Math
>            Reporter: Ted Dunning
>            Assignee: Ted Dunning
>             Fix For: 0.5
>
>         Attachments: MAHOUT-376.patch, Modified stochastic svd algorithm for mapreduce.pdf, QR decomposition for Map.pdf, sd-bib.bib, sd.pdf, sd.pdf, sd.pdf, sd.pdf, sd.tex, sd.tex, sd.tex, sd.tex, Stochastic SVD using eigensolver trick.pdf
>
>
> See attached pdf for outline of proposed method.
> All comments are welcome.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.