[
https://issues.apache.org/jira/browse/MAHOUT-796?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13091532#comment-13091532
]
Dmitriy Lyubimov commented on MAHOUT-796:
-----------------------------------------
Orthogonalization, i wouldn't say it has been a bottleneck for me personally. I
do orthogonalization in two MR steps, but those steps are also doing 2
multiplications. I am still looking to run an experiment with an input where it
would be a practical problem (such as QR step taking > 10..15 min). It also so
happened that my previous results were tainted by the fact that i ran without
native libraries installed so decompression was chomping on my CPU too, so no
clean data now. But it seems indeed like roughly 50% of everything else.
but i am still waiting for run-a-terabyte-input opportunity.
> 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
