LR in GSoC.Mahout

[Yun Jiang <he...@gmail.com>]

Dears,

I'm a junior student, and I would like to apply for Logistic Regression in
Mahout. I've read the NIPS paper, and I hope can implement the LR for both
binary class and Multiclass. How do you think?

Best regards,
Yun Jiang

Re: LR in GSoC.Mahout

[Isabel Drost <ap...@isabel-drost.de>]

On Saturday 29 March 2008, Ted Dunning wrote:
> The basic outline is to set up a Jira request for enhancement that
> describes what you want to do, write or find a sequential version for
> reference and then start on the actual coding.

If you want to get funding for your project from Google, you might in addition 
add an application at the GSoC web application.

Isabel

-- 
Systems programmers are the high priests of a low cult.		-- R. S. Barton
  |\      _,,,---,,_       Web:   <http://www.isabel-drost.de>
  /,`.-'`'    -.  ;-;;,_
 |,4-  ) )-,_..;\ (  `'-'
'---''(_/--'  `-'\_) (fL)  IM:  <xm...@spaceboyz.net>

Re: LR in GSoC.Mahout

[Ted Dunning <td...@veoh.com>]

That sounds great.

Take a look at the discussion archives to get a feel for how people have
been contributing code.

The basic outline is to set up a Jira request for enhancement that describes
what you want to do, write or find a sequential version for reference and
then start on the actual coding.

As a reference for the actual algorithms to use, Andrew Gelman has a very
good article with associated R code here:
http://www.stat.columbia.edu/~gelman/standardize/

Note also that logistic regression is a specialization of a family of
techniques called generalized linear modeling.  If you play your cards
right, you can implement logistic, probit, Poisson and various kinds of
linear regression all in one step.  If you use the outline from Andrew's
code, then you also will get a Bayesian framework for it as well which can
be used for all kinds of interesting add-ons.


On 3/29/08 8:05 AM, "Yun Jiang" <he...@gmail.com> wrote:

> Dears,
> 
> I'm a junior student, and I would like to apply for Logistic Regression in
> Mahout. I've read the NIPS paper, and I hope can implement the LR for both
> binary class and Multiclass. How do you think?
> 
> Best regards,
> Yun Jiang