SVM kernel error

[igor gregovitch]

Hi,

I'm using Rattle with debian, and I've got alway this error, whatever
I update packages, when a try svm model :
"The call to svm appears to have failed. The error message was: Error
in svm.default(x, y, scale = scale, ..., na.action = na.action) :
wrong kernel specification!
I am not familiar with this error, and you may want to report it to
at sup...@togaware.com"
The log is :
crs$svm <- svm(BQ2_myvar ~ ., data=crs$dataset[crs$sample,c(1:12,14)],
kernel="rbfdot", probability=TRUE)
When I try with ksvm instead of svm in console, it works, but it
doesn't update de Rattle windows.
What's wrong with it ?

Thanks.

(Whatever, Rattle is great)

[Richard Palmer]

I too have got the same msg.  Further I agree that rattle is great.

--
Richard Palmer

Home 508 877-3862
Cell 508 982-7266

[Graham Williams]

Hi Igor, Richard,

What version of Rattle are you using?

I've tried repeating this in the current beta version (2.3.37) and
don't see the same problem (and it is using ksvm rather than
svm).

Also, Igor, after you do your own "crs$svm <- ..." in the console,
whilst the Rattle textview for the SVM won't be updated, if you
"Evaluate" the SVM in Rattle, it should work okay, using your model.

But of course, we should fix the problem!

Regards,
Graham

Received Tue 17 Jun 2008 5:07am +1000 from igor gregovitch:

[Richard Palmer]

I downloaded the beta after your last announcement.  My problem occured on the earlier production versioin (I think).  Will try to reproduce..

[igor gregovitch]

Hi,

I'm using the version 2.3.37, but I've this error on earlier version
too. If I redo the command in R, I've got same error :
"Erreur dans svm.default(x, y, scale = scale, ..., na.action =
na.action) :
wrong kernel specification!" (I'm not french, I swear)
Just if I replace svm by ksvm it seems to work in R, cause R say :
Using automatic sigma estimation (sigest) for RBF or laplace kernel
but in Rattle then, when I try to evaluate SVM :
E120: Some model has not been built? We found the model types to be:
ksvm The models not built: TRUE This is a Rattle bug. Please report
this to sup...@togaware.com.
In target, I use a categorical 2 variable. But with a numeric boolean,
same thing. Others models works fine (neural net is great, thanks).

On 18 juin, 03:26, "Richard Palmer" <rhpal...@gmail.com> wrote:
> I downloaded the beta after your last announcement. My problem occured on
> the earlier production versioin (I think). Will try to reproduce..
>
> On Tue, Jun 17, 2008 at 3:47 PM, Graham Williams <
>
>
>
> Graham.Willi...@togaware.com> wrote:
>
> > Hi Igor, Richard,
>
> > What version of Rattle are you using?
>
> > I've tried repeating this in the current beta version (2.3.37) and
> > don't see the same problem (and it is using ksvm rather than
> > svm).
>
> > Also, Igor, after you do your own "crs$svm <- ..." in the console,
> > whilst the Rattle textview for the SVM won't be updated, if you
> > "Evaluate" the SVM in Rattle, it should work okay, using your model.
>
> > But of course, we should fix the problem!
>
> > Regards,
> > Graham
>
> > Received Tue 17 Jun 2008 5:07am +1000 from igor gregovitch:
>
> > > Hi,
>
> > > I'm using Rattle with debian, and I've got alway this error, whatever
> > > I update packages, when a try svm model :
> > > "The call to svm appears to have failed. The error message was: Error
> > > in svm.default(x, y, scale = scale, ..., na.action = na.action) :
> > > wrong kernel specification!
> > > I am not familiar with this error, and you may want to report it to

> > > at supp...@togaware.com"

[igor gregovitch]

Hi,

It works fine with the 2.3.47 version, after update or install some
package (under debian : r-cran-lme4 r-cran-matchit r-cran-mcmcpack r-
cran-mvtnorm r-cran-sandwich r-cran-zelig r-other-gking-matchit r-
base r-base-core r-cran-matrix r-recommended ) (well, one of them must
be the key). SVM works fine. Thanks.




(

On 18 juin, 22:00, igor gregovitch <d.bal...@gmail.com> wrote:
> Hi,
>
> I'm using the version 2.3.37, but I've this error on earlier version
> too. If I redo the command in R, I've got same error :
> "Erreur dans svm.default(x, y, scale = scale, ..., na.action =
> na.action) :
> wrong kernel specification!" (I'm not french, I swear)
> Just if I replace svm by ksvm it seems to work in R, cause R say :
> Using automatic sigma estimation (sigest) for RBF or laplace kernel
> but in Rattle then, when I try to evaluate SVM :
> E120: Some model has not been built? We found the model types to be:
> ksvm The models not built: TRUE This is a Rattle bug. Please report

> this to supp...@togaware.com.

[Graham Williams]

Thanks for updating and trying that out Igor. Good to see it is now
working.

Regards,
Graham

Received Tue 08 Jul 2008 3:16am +1000 from igor gregovitch:

[Tony Nolan]

GOOD Morning,

Is there a COP today ?

Any talk time too ?

Regards
Tony

--
UTS CRICOS Provider Code: 00099F
DISCLAIMER: This email message and any accompanying attachments may contain
confidential information. If you are not the intended recipient, do not
read, use, disseminate, distribute or copy this message or attachments. If
you have received this message in error, please notify the sender immediately
and delete this message. Any views expressed in this message are those of the
individual sender, except where the sender expressly, and with authority,
states them to be the views of the University of Technology Sydney. Before
opening any attachments, please check them for viruses and defects.

Think. Green. Do.

Please consider the environment before printing this email.

[Graham Williams]

Hi Tony,

No CoP today. We are having a local afternoon tea.

Regards,
Graham

Received Tue 08 Jul 2008 6:37am +1000 from Tony Nolan:

[Graham Williams]

Ooops - sorry for the "Spam" on the mailing list......

Regards,
Graham

Received Tue 08 Jul 2008 6:58am +1000 from Graham Williams: