Hi, Roberto > Does anyone know if rbf is constrained in terms of number of dimensions?
No, there is no constraint on the number of dimensions > I sometimes get weird results with SVM and rbf kernel in terms of false > positive/negative rates. Maybe you are overfitting or the data is just not separable even in the infinite RBF space. The fine tuning of gamma is really crucial, however, sometimes there may not be a gamma value that yields satisfactory results. You can also try to use other kernels instead of RBF. Based on your first question, it sounds like that you are working with high-dimensional data. Maybe it would be worthwhile to either do feature selection or dimensionality reduction to avoid overfitting. Since I assume that you have a non-linear problem here (because you are using a kernel SVM), I would suggest to try if kernel PCA or kernel LDA for prior dim. red. can improve your performance. The scikit-learn implementation would be from sklearn.decomposition import KernelPCA scikit_kpca = KernelPCA(n_components=2, kernel='rbf', gamma=15) X_skernpca = scikit_kpca.fit_transform(X) If you are interested, I have a stepwise tutorial for kernel PCA to illustrate the concept a little bit: http://sebastianraschka.com/Articles/2014_kernel_pca.html Good luck! Sebastian > On Oct 21, 2014, at 8:58 AM, Pagliari, Roberto <rpagli...@appcomsci.com> > wrote: > > I sometimes get weird results with SVM and rbf kernel in terms of false > positive/negative rates. > > I suspect there may be numerical issues going on, because I’m not seeing the > same issues with linearSVC. > > Does anyone know if rbf is constrained in terms of number of dimensions? > > Unfortunately I cannot share the data I am using. > > Thank you, > > ------------------------------------------------------------------------------ > Comprehensive Server Monitoring with Site24x7. > Monitor 10 servers for $9/Month. > Get alerted through email, SMS, voice calls or mobile push notifications. > Take corrective actions from your mobile device. > http://p.sf.net/sfu/Zoho_______________________________________________ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general ------------------------------------------------------------------------------ Comprehensive Server Monitoring with Site24x7. Monitor 10 servers for $9/Month. Get alerted through email, SMS, voice calls or mobile push notifications. Take corrective actions from your mobile device. http://p.sf.net/sfu/Zoho _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
