Hi Mathieu,
yes, in 2.0 we changed the way algorithms are written. It is now much simpler for the user but not compatible with the old 1.3 release. The development releases do not contain yet all the algorithms that were available before.
I’m attaching the sda_featex.m updated for 2.0 to this message. It will appear in the stable 2.1 release coming soon.
You may also check out other algorithm examples in the distribution, namely the sda_pca_clf.m and sda_pca_roc.m. The first one trains a classifier in PCA-reduced subspace and uses the default operating point (equal weights for all classes). The second one also sets the operating point using ROC analysis.
Pavel
