Hello, The previous email, mentioned PCA and ICA and their kernel counter parts. Like it sauid, its a good place to start from. But their effectiveness highly depends on how noisy your dataset is and how large is the number of features.
SVM-RFE by Isabelle Guyon is known to be very good. Another method is mentioned in 'Feature Selection for High-Dimensional Genomic Microarray Data' by eric xing et al. In the paper titled 'Feature selection methods in kernel based gene expression analysis' in Kernel Methods in Computational Biology, B. Scholkopf, K. Tsuda, and J.-P. Vert (editors), MIT press, 2004 would give you a good overview of the categories and examples of feature selection methods. Many of these algorithms can be found implemented in matlab ..and probably even in Java/C. Thank You. Sincerely, Monika Ray *********************************************************************** The sweetest songs are those that tell of our saddest thought... Computational Intelligence Centre, Washington University St. louis, MO ********************************************************************** On Wed, 25 May 2005, Felix Goldberg wrote: > Well, you can use PCA which is a standard tool. (Of course, depending on > your problem and the nature and quality of the features the results may range > from great to catastrophic). In any case, that's what I'd try first. > > Another option would be ICA but you need to satisfy some additional > statistical > assumptions for that to work. Worth giving it a try anyway. And there are > kernel versions of PCA and ICA around too. > > You haven't specified in what language you are working so I'll suggest > implementations for MATLAB with which I'm best familiar: (i) PCA is > built-in in (at least the recent versions of) MATLAB and you can also > google for the very nice free toolbox called SPRTool. (ii) for ICA you can > google for the free toolbox FastICA. > > If you wish to discuss the question further feel free to mail me. > > With best regards, > Felix. > > --------------------------------------- > Felix Goldberg, M.Sc. > [EMAIL PROTECTED] > www.technion.ac.il/~felixg > > On Wed, 25 May 2005, Estevam Rafael Hruschka Junior wrote: > > > Dear all, > > > > I'm performing a clustering task and need to reduce the amount of my dataset > > features (attributes). Looking for a feature selection tool, I've only faced > > packages for classification. As I do not have the class information (in my > > dataset) I am not able to use them. > > > > Do you know about any (free) feature selection software for clustering > > tasks? > > > > > > > > Thank you so much in advance. > > > > Best regards, > > > > > > > > Estevam. > > > > -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_- > > Estevam Rafael Hruschka Junior, PhD > > Departamento de Computação > > Universidade Federal de São Carlos (DC-UFSCar) > > Rod. Washington Luiz, km 235 > > Caixa Postal (PO Box) 676 > > 13565-905 São Carlos - SP - Brazil > > phone: +55 16 3351-8608 > > http://www.dc.ufscar.br/~estevam > > > > _______________________________________________ > > uai mailing list > > uai@ENGR.ORST.EDU > > https://secure.engr.oregonstate.edu/mailman/listinfo/uai > > _______________________________________________ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
