You mis-use map-reduce is how you do it.

Start a bunch of cores and nominate some of them to be parameter stores and
many to be workers.  Pretend you are running either a map-only task or lots
of reducers on an identity mapper.

On Thu, Mar 14, 2013 at 12:00 PM, Hector Yee <[email protected]> wrote:

> How do you do 'asynchronous sgd" with hadoop? (I guess referring to the
> spark discussion in a different thread)
>
> In any case there are ways to get similar performance without having to use
> that many cores.
>
> See our paper 'affinity weighted embedding"
>
> http://openreview.net/iclr2013
>
> table 2, compared to the result you mentioned
>
> http://arxiv.org/pdf/1301.4171v1.pdf
>
>
> On Thu, Mar 14, 2013 at 11:44 AM, Ted Dunning <[email protected]>
> wrote:
>
> > Yeah.... we have had little pull on these techniques beyond the simplest
> > case of logistic regression.
> >
> > Would you guys be willing to sign up for maintaining the code that might
> > result?
> >
> > The thing that might move the needle would be a replication of this
> > architecture:
> >
> >
> >
> http://deeplearning.net/2012/12/13/googles-large-scale-deep-learning-experiments/
> >
> > On Thu, Mar 14, 2013 at 8:33 AM, Danny Busch <[email protected]> wrote:
> >
> > > I asked years ago whether someone would be interested in starting
> > >  implementations of neural network algorithms in mahout, but
> apparently t
> > > here
> > > was no big interest / priority and feedback that time.  Maybe this has
> > > changed,
> > > as Neural Networks yet again become more popular.
> > >
> > > ~ danny
> > >
> > >
> > > Ying Liao <[email protected]> hat am 14. März 2013 um 16:24
> > geschrieben:
> > > > Can anyone in dev/commits shed some lights on this?
> > > >
> > > > Thanks,
> > > > Ying
> > > >
> > > > ---------- Forwarded message ----------
> > > > From: Ying Liao <[email protected]>
> > > > Date: Wed, Mar 13, 2013 at 6:04 PM
> > > > Subject: Neural Network and Restricted Boltzman Machine in Mahout
> > > > To: [email protected]
> > > >
> > > >
> > > > Hi,
> > > >
> > > > I am interested using Neural Network type of tools to do
> classification
> > > in
> > > > Mahout. I wonder if these methods are finished in Mahout 0.8. On
> Mahout
> > > > website, it still says open.
> > > >
> > > > Thanks,
> > > > Ying
> > >
> >
>
>
>
> --
> Yee Yang Li Hector <https://plus.google.com/106746796711269457249>
> Professional Profile <http://www.linkedin.com/in/yeehector>
> http://hectorgon.blogspot.com/ (tech + travel)
> http://hectorgon.com (book reviews)
>

Reply via email to