mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nowal, Akshay" <Akshay_No...@SYNTELINC.COM>
Subject RE: Support Vector Machine in Mahout
Date Mon, 02 Jul 2012 04:36:24 GMT
Hi Ted,

Thanks for the quick reply.

Actually m new in using Mahout and always use trunk for running the algos.
I don't have much knowledge of Java. So is there any command through trunk that can do this?

(But how large is your data in any case?  Do you actually need a parallelized algorithm?)
The data is in millions of records, other data has millions of comments that are to be classified
and it has to b updated as n when new comments are received. And we want to showcase the advantage
of parallel processing also so was thinking if it's available?

Regards,
Akshay Nowal

 |       

-----Original Message-----
From: Ted Dunning [mailto:ted.dunning@gmail.com] 
Sent: Friday, June 29, 2012 6:46 PM
To: user@mahout.apache.org
Subject: Re: Support Vector Machine in Mahout

On Fri, Jun 29, 2012 at 1:13 AM, Nowal, Akshay
<Akshay_Nowal@syntelinc.com>wrote:

>
> I am at a beginner level in using Mahout and m planning to build a
> classifier on Customer data  to classify churners and non-churners using
> support vector machine(SVM).
>

The easiest way to do this is to add a hinge-loss variant to the SGD
algorithm already in Mahout (see OnlineLogisticRegression for an example
using logistic loss).


>
> Currently does any parallelized algorithm SVM is available?
>

Not currently.

But how large is your data in any case?  Do you actually need a
parallelized algorithm?
Mime
View raw message