On 6/7/09 9:51 AM, "Carlos A. Garcma Vallejo" wrote:
> Hello (sorry for my poor English).
>
> I have a question about CFS.
>
> Hello. I have a question about CFS. The first thing I want to say is
> that I feel like you have a question about The Godfather and write a
> list in which Francis Ford Coppola answers.
I don't feel like the Godfather - I haven't had any severed horses heads placed
in anyone's bed (at least not recently) :-)
>
> Well, I am writting a new (and hope that good) Feature Selection
> Algorithm. It's based on the correlation between features, so,
> obviously, I use CFS (!thanks Mark!). I have modified the class
> CFSSubsetEval, calling MyCFSSubsetEval, making the field m_corr_matrix
> public. Then I write the following code:
>
> MyCfsSubsetEval cfs = new MyCfsSubsetEval();
> int numFeaturesIncludedClass = instances.numAttributes();
> numFeatures = numFeaturesIncludedClass - 1;
> BitSet bs = new BitSet(numFeatures);
> for (int i = 0; i < numFeatures; i++) {
> bs.set(i);
> }
> cfs.buildEvaluator(instances);
> cfs.evaluateSubset(bs);
> float[][] correlationMatrix = cfs.m_corr_matrix;
>
> and I do have the correlations.
>
> The question is: is this the best way I could do it, or there is another
> more efficient way?
This looks OK to me. I should probably make m_corr_matrix protected and add an
accessor method for it. It didn't occur to me at the time that anyone might want
to subclass CFS.
Cheers,
Mark.
--
Mark Hall
Senior Developer/Consultant, Pentaho Open Source Business Intelligence
Citadel International, Suite 340, 5950 Hazeltine National Dr., Orlando, FL
32822, USA
+64 7 847-3537 office, +64 21 399-132 mobile, +1 815 550-8637 fax,
Skype: mark.andrew.hall, Yahoo: mark_andrew_hall
Download the latest release today <
http://www.sourceforge.net/projects/pentaho>
_______________________________________________
Wekalist mailing list
Send posts to:
Wekalist@...
List info and subscription status:
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalistList etiquette:
http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html