[
https://issues.apache.org/jira/browse/MAHOUT-796?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13091533#comment-13091533
]
Nathan Halko commented on MAHOUT-796:
-------------------------------------
We did some work with facial recognition, computing 'eigenfaces' and reported
in the paper. In this case there is only 2 orders of magnitude between the
signal and the 'noise'. It shows a dramatic difference between the accuracy of
one pass versus just one power iteration. Note that after one power iteration,
there is now 6 orders of magnitude separating signal and noise.
But this is only looking at approximation error ||A-UU*A||. It could very well
be the case in recommendation applications that this measure is not
appropriate, I don't know. But it is a very valuable option to have at one's
disposal just in case.
> Modified power iterations in existing SSVD code
> -----------------------------------------------
>
> Key: MAHOUT-796
> URL: https://issues.apache.org/jira/browse/MAHOUT-796
> Project: Mahout
> Issue Type: Improvement
> Components: Math
> Affects Versions: 0.5
> Reporter: Dmitriy Lyubimov
> Assignee: Dmitriy Lyubimov
> Labels: SSVD
> Fix For: 0.6
>
>
> Nathan Halko contacted me and pointed out importance of availability of power
> iterations and their significant effect on accuracy of smaller eigenvalues
> and noise attenuation.
> Essentially, we would like to introduce yet another job parameter, q, that
> governs amount of optional power iterations. The suggestion how to modify the
> algorithm is outlined here :
> https://github.com/dlyubimov/ssvd-lsi/wiki/Power-iterations-scratchpad .
> Note that it is different from original power iterations formula in the paper
> in the sense that additional orthogonalization performed after each
> iteration. Nathan points out that that improves errors in smaller eigenvalues
> a lot (If i interpret it right).
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira