SVM w/ radial basis function

[yonestar]

Using 'rbf' flag in 'cv_svm' for predict, the predicted values (stats.yfit) is all 1's. That is, it predicts all observations as belonging to ONE class.

Without the radial basis function, some observations are predicted to be class A, others as class B, as expected.

The file for reproducing the error is in the repo, see CanlabPrivate/FILES_FOR_REPRODUCING_BUGS/svm_rbf_bug

[ljchang]

Hmm...I think we have changed some things to how the predict function works after this was implemented and I doubt anyone has used it since. I think we decided that the nonlinear algorithms weren't terribly interpretable when we were playing with this in Tor's class a few years ago.