SVM question

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

SVM question

Georges Orlowski
I'm running SVM from e1071 package on a data with ~150 columns (variables)
and 50000 lines of data (it takes a bit of time) for radial kernel for
different gamma and cost values.

I get a very large models with at least
30000 vectors and the prediction I get is not the best one. What does it
mean and what could I do to ameliorate my model ?

Jerzy Orlowski

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Reply | Threaded
Open this post in threaded view
|

Re: SVM question

Uwe Ligges
Georges Orlowski wrote:

> I'm running SVM from e1071 package on a data with ~150 columns (variables)
> and 50000 lines of data (it takes a bit of time) for radial kernel for
> different gamma and cost values.
>
> I get a very large models with at least
> 30000 vectors and the prediction I get is not the best one. What does it
> mean and what could I do to ameliorate my model ?

Do you mean 30000 *support vectors* in 50000 observations? So you are
heavily overfitting. Try to tune the svm better.

Uwe Ligges


>
> Jerzy Orlowski
>
> ______________________________________________
> [hidden email] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Reply | Threaded
Open this post in threaded view
|

Re: SVM question

Georges Orlowski
Well, I know there must be something wrong with the way I'm running the
SVM but I tried all the posiible ranges of parameters gamma and cost and
it does not improve? Could You suggest anything?


On Wed, 1 Feb 2006, Uwe Ligges wrote:

> Georges Orlowski wrote:
>
> > I'm running SVM from e1071 package on a data with ~150 columns (variables)
> > and 50000 lines of data (it takes a bit of time) for radial kernel for
> > different gamma and cost values.
> >
> > I get a very large models with at least
> > 30000 vectors and the prediction I get is not the best one. What does it
> > mean and what could I do to ameliorate my model ?
>
> Do you mean 30000 *support vectors* in 50000 observations? So you are
> heavily overfitting. Try to tune the svm better.>
> Uwe Ligges
>
>
> >
> > Jerzy Orlowski
> >
> > ______________________________________________
> > [hidden email] mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Reply | Threaded
Open this post in threaded view
|

Re: SVM question

Liaw, Andy
In reply to this post by Georges Orlowski
Have you tried the range of parameters suggested in

  http://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf

?

Please do read the posting guide to help you phrase the question better.
Without showing us exactly what you did, we have no way of telling what
might have gone wrong.  The only information you gave us is that you tried
SVM on your data and it didn't do so well.  What is anyone supposed to do
with that info?

Andy

From: Georges Orlowski
 

>
> Well, I know there must be something wrong with the way I'm
> running the
> SVM but I tried all the posiible ranges of parameters gamma
> and cost and
> it does not improve? Could You suggest anything?
>
>
> On Wed, 1 Feb 2006, Uwe Ligges wrote:
>
> > Georges Orlowski wrote:
> >
> > > I'm running SVM from e1071 package on a data with ~150
> columns (variables)
> > > and 50000 lines of data (it takes a bit of time) for
> radial kernel for
> > > different gamma and cost values.
> > >
> > > I get a very large models with at least
> > > 30000 vectors and the prediction I get is not the best
> one. What does it
> > > mean and what could I do to ameliorate my model ?
> >
> > Do you mean 30000 *support vectors* in 50000 observations?
> So you are
> > heavily overfitting. Try to tune the svm better.>
> > Uwe Ligges
> >
> >
> > >
> > > Jerzy Orlowski
> > >
> > > ______________________________________________
> > > [hidden email] mailing list
> > > https://stat.ethz.ch/mailman/listinfo/r-help
> > > PLEASE do read the posting guide!
> http://www.R-project.org/posting-guide.html
> >
>
> ______________________________________________
> [hidden email] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
> http://www.R-project.org/posting-guide.html
>
>

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html