SMO turn off normalization

_______________________________________________
Wekalist mailing list
Wekalist@...
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

> On page 410 of weka's book (Data Mining…by Witten and Frank), they suggest
> to turn normalization off for faster operation of SMO when working with
> sparse instances.
>
> How do I turn normalization off when executing SMO from the command line?

With the book version of Weka you can't, with the developer version
use "-no-checks".

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174


_______________________________________________
Wekalist mailing list
Wekalist@...
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

I used "-no-checks" as follows:
   java -Xmx6000m weka.classifiers.functions.SMO -no-checks -C 1.0 -L 0.0010 -P 1.0E-12 -N 0 -V -1 -W 1 -K "weka.classifiers.functions.supportVector.PolyKernel -C 250007 -E 1.0" -t Train.arff -T Test.arff -i -p 0

And got the following error:
weka.core.UnsupportedAttributeTypeException: weka.classifiers.functions.supportV
ector.PolyKernel: Cannot handle binary attributes!
        at weka.core.Capabilities.test(Unknown Source)
        at weka.core.Capabilities.test(Unknown Source)
        at weka.core.Capabilities.test(Unknown Source)
        at weka.core.Capabilities.test(Unknown Source)
        at weka.core.Capabilities.testWithFail(Unknown Source)
        at weka.classifiers.functions.supportVector.CachedKernel.buildKernel(Unk
nown Source)
        at weka.classifiers.functions.SMO$BinarySMO.buildClassifier(Unknown Sour
ce)
        at weka.classifiers.functions.SMO.buildClassifier(Unknown Source)
        at weka.classifiers.Evaluation.evaluateModel(Unknown Source)
        at weka.classifiers.Classifier.runClassifier(Unknown Source)
        at weka.classifiers.functions.SMO.main(Unknown Source)

I was able to run successfully the classifier without the "-no-checks" option.

Thanks,

Martha L.
-----Original Message-----
From: wekalist-bounces@... [mailto:wekalist-bounces@...] On Behalf Of Peter Reutemann
Sent: Wednesday, June 25, 2008 5:27 PM
To: Weka machine learning workbench list.
Subject: Re: [Wekalist] SMO turn off normalization

> On page 410 of weka's book (Data Mining...by Witten and Frank), they
> suggest to turn normalization off for faster operation of SMO when
> working with sparse instances.
>
> How do I turn normalization off when executing SMO from the command line?

With the book version of Weka you can't, with the developer version use "-no-checks".

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174



_______________________________________________
Wekalist mailing list
Wekalist@...
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

The kernels have a "-no-check" option as well. But it looks like
you're trying to use data that is *not* suitable for SMO. The data
needs to be all numeric attributes, apart from the class attribute
being nominal. Using the the "-no-checks" option is highly dangerous
as it might create completely useless results if you're supplying the
wrong data type (with that option present, SMO performs *no* data
transformations - you have to do it yourself).

Cheers, Peter
--
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
http://www.cs.waikato.ac.nz/~fracpete/ Ph. +64 (7) 858-5174

_______________________________________________
Wekalist mailing list
Wekalist@...
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist