You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by Josh Patterson <jo...@cloudera.com> on 2011/05/28 00:54:49 UTC

SGD Performance

Just wanted to say I really enjoyed having dinner with everyone last
night at the Mahout dinner in PA;

Ted mentioned I believe (sequential) SGD performance last night as
being pretty incredible in terms of overall training time especially
compared to some other MR based techniques (If I have this wrong, Ted,
please correct me) --

I was talking about this with a coworker this morning; Are there any
online published stats comparing this in hard numbers?

Thanks,

JP

-- 
Twitter: @jpatanooga
Solution Architect @ Cloudera
hadoop: http://www.cloudera.com
blog: http://jpatterson.floe.tv

Re: SGD Performance

Posted by Hector Yee <he...@gmail.com>.
We used a lot of sgd as opposed to Mr at google see papers. By Samy bengio
On May 27, 2011 5:55 PM, "Josh Patterson" <jo...@cloudera.com> wrote:
> Just wanted to say I really enjoyed having dinner with everyone last
> night at the Mahout dinner in PA;
>
> Ted mentioned I believe (sequential) SGD performance last night as
> being pretty incredible in terms of overall training time especially
> compared to some other MR based techniques (If I have this wrong, Ted,
> please correct me) --
>
> I was talking about this with a coworker this morning; Are there any
> online published stats comparing this in hard numbers?
>
> Thanks,
>
> JP
>
> --
> Twitter: @jpatanooga
> Solution Architect @ Cloudera
> hadoop: http://www.cloudera.com
> blog: http://jpatterson.floe.tv

Re: SGD Performance

Posted by Ted Dunning <te...@gmail.com>.
Start here: http://leon.bottou.org/projects/sgd

Then here: http://www.cs.jhu.edu/~mdredze/publications/icml_variance.pdf

On Fri, May 27, 2011 at 3:54 PM, Josh Patterson <jo...@cloudera.com> wrote:

> Ted mentioned I believe (sequential) SGD performance last night as
> being pretty incredible in terms of overall training time especially
> compared to some other MR based techniques (If I have this wrong, Ted,
> please correct me) --
>
> I was talking about this with a coworker this morning; Are there any
> online published stats comparing this in hard numbers?
>