Richard Kirkby | 11 Apr 23:42 2002
Picon
Picon

Re: Class Value

Hi Wendy,

distributionForInstance is not a good place to do this. The evaluation
modules in Weka will deliberately set the class of test instances to
missing to remove any possibility of cheating.
As for suggesting a better spot, I'm not really sure, but it would have
to be some part of code where you have control over the test instances
so as to get at the class values that you need.

Richard
Weka support
--

Wendy wrote:
> 
> hi,
>         I tried using classValue() too.  The program says that the class
> value is missing.  What I am trying to do is to record the top 20 NN
> output values and their classes.  Currently I am trying to get these
> values in distributionForInstance.  I don't think that this is the correct
> location to do this but I could not find a better location. Is their
> another spot that would not always have missing class values?
> 
> Thank you very much for you help,
> Wendy
> 
> On Thu, 11 Apr 2002, Richard Kirkby wrote:
> 
> > Hi Wendy,
> >
(Continue reading)

Perfecto Herrera | 12 Apr 17:53 2002
Picon

default folder

Is there any way of defining the default folder for ARFF files in the
Windows version?

Thank you very much.

--

-- 
Perfecto Herrera Boyer                          
Institut de l'Audiovisual - Universitat Pompeu Fabra  
Pg. Circumval.lació 8, 08003 Barcelona                     
Spain                                       

phone 34 93 542 28 06
fax   34 93 542 22 02 

----
Once the search is in progress something will be found
- Brian Eno, Oblique Strategies
Erik Geleyn | 12 Apr 19:35 2002

ARFF format

Hi,

I am sorry, I lost the email with the link of the description of the ARFF
format. Could someone please send it to me?

Thanks
Erik

*****************************************
Erik Geleyn
Grad Research Asst-Ops 
Florida Atlantic University
egeleyn <at> cse.fau.edu
egeleyn <at> hotmail.com
Homepage: www.cse.fau.edu/~egeleyn
(561) 750 9258
(561) 297 2512
*****************************************
David A. Vavra | 13 Apr 20:54 2002
Picon

RE: default folder

I created a shortcut.

  Target: <Weka home folder>\weka.jar
  Start in: <desired data folder>

Click on the shortcut to start Weka

DAV

> -----Original Message-----
> From: wekalist-admin <at> orc.cs.waikato.ac.nz
> [mailto:wekalist-admin <at> orc.cs.waikato.ac.nz]On Behalf Of Perfecto
> Herrera
> Sent: Friday, April 12, 2002 11:53 AM
> To: wekalist <at> orc.cs.waikato.ac.nz
> Subject: [Wekalist] default folder
>
>
> Is there any way of defining the default folder for ARFF files in the
> Windows version?
>
> Thank you very much.
>
>
> --
> Perfecto Herrera Boyer
> Institut de l'Audiovisual - Universitat Pompeu Fabra
> Pg. Circumval.lació 8, 08003 Barcelona
> Spain
>
(Continue reading)

Richard Kirkby | 14 Apr 23:44 2002
Picon
Picon

Re: ARFF format

Hi,

The description of the ARFF format, kindly put together by Gordon
Paynter, is now linked off the Weka site. Here is the link:

http://www.cs.waikato.ac.nz/~ml/weka/arff.html

Cheers,

Richard
Weka support
--

Erik Geleyn wrote:
> 
> Hi,
> 
> I am sorry, I lost the email with the link of the description of the ARFF
> format. Could someone please send it to me?
> 
> Thanks
> Erik
> 
> *****************************************
> Erik Geleyn
> Grad Research Asst-Ops
> Florida Atlantic University
> egeleyn <at> cse.fau.edu
> egeleyn <at> hotmail.com
> Homepage: www.cse.fau.edu/~egeleyn
(Continue reading)

Josh Anderson | 17 Apr 02:22 2002
Picon

retrieving classifier predictions when using cross validation

Is there any way to retrieve the actual classifier predictions for each instance, when the classifier is implemented with 10 fold cross validation?  In the tutorial, it said that the only way to retrieve the actual classifier predictions involves providing a seperate test set, and I wanted to make sure that that was accurate.
 
thanks,
 
Josh Anderson
Yu Wang | 18 Apr 15:40 2002

SMO parameter

Hello, I wonder if anybody has implemented the Gaussian Kernel for SMO? 
(If you search in SVM literature, most papers deal with the Gaussian 
Kernel.) Second, what does "use the lower order terms" mean? Is it for 
the kernel like this "(<X,Y>+1)^d"?

Thx,

YW

--

-- 
Yu Wang                                          http://www.PharmaDM.com 
PharmaDM NV.                                     TEL: +32-16-387480 
Ambachtenlaan 54/D, B-3001 Leuven, Belgium       Fax: +32-16-387471
Eibe Frank | 18 Apr 23:10 2002
Picon
Picon

Re: SMO parameter


> Second, what does "use the lower order terms" mean? Is it for
> the kernel like this "(<X,Y>+1)^d"?

Correct.

Cheers,
Eibe
Benjamin Good | 22 Apr 20:28 2002

string kernels?

Has anyone looked into adding a string kernel to the support vector machine
implementation in WEKA?  Could be very useful for molecular biology and text
classification applications.

thanks
-Ben
Aldebaro Klautau | 22 Apr 21:09 2002

making available RBF kernel to Weka's SVM

Some people asked about RBF (Gaussian) kernel for Weka's SVM. I am sending
attached a modified SMO code, which implements RBF and polynomial kernels
for non-sparse instances (a bit faster than Weka's original implementation).
NOTE: The new kernels assume the class is the last attribute (Weka's
original kernel is the default option, and doesn't have this restriction).
It's code I use for my research and it may have bugs, but you can freely use
it (or incorporate to the Weka package - in this case I should go over the
code and update its documentation).

Aldebaro

==============================
Aldebaro Klautau
UCSD - ECE - Ph.D. candidate
http://speech.ucsd.edu/aldebaro

----- Original Message -----
From: "Benjamin Good" <bgood <at> vapop.ucsd.edu>
To: "weka list" <wekalist <at> orc.cs.waikato.ac.nz>
Sent: Monday, April 22, 2002 11:28 AM
Subject: [Wekalist] string kernels?

> Has anyone looked into adding a string kernel to the support vector
machine
> implementation in WEKA?  Could be very useful for molecular biology and
text
> classification applications.
>
> thanks
> -Ben

Attachment (SMO.java): text/java, 50 KiB

Gmane