Just an update.

I am still working out a version that would do 100% mapreduce version of Q
orhonormalization . I think i made some progress with the prototype, and
basically there seems also to be a way to drastically cut on the output of
the first map reduce (i.e. reduce amount of partial B^t outer products)
depending on how much one wants to compromise the stochastic effect produced
by Y. The algorithm also significantly simplifies the orthogonolization
process by basically getting rid of Y-blocking entirely inside the same
mapper (i.e. it's 100% streaming, 1 Q-block per mapper with memory
requirements still constant to size of a row of A).

It's slow due to my family situtation (i had just several hours to work on
non-mr prototype so far, which passed Q orthogonality test, everything else
is just a thought experiment)

On Sun, Oct 24, 2010 at 12:11 AM, Ted Dunning (JIRA) <[email protected]>wrote:

>
>     [
> https://issues.apache.org/jira/browse/MAHOUT-376?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel]
>
> Ted Dunning updated MAHOUT-376:
> -------------------------------
>
>    Attachment: sd.tex
>                sd.pdf
>
> Updated version.
>
> I think that this is actually a feasible algorithm.
>
> > 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, 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.
>
>

Reply via email to