On Tue, Mar 06, 2012 at 09:52:59PM +0100, Andreas wrote: > Are you talking about coordinate descent in the primal or in the dual?
Once again, I am not an expert, but I believe that for l1-penalized logistic regression with n_features >> n_samples, coordinate descent in the primal is faster. > My main interest in this was having a SGD solver for Crammer-Singer > loss, which works nearly the same way as logistic regression. Yes, I think that SGD answer a different situation and that we should also have one. Gaël ------------------------------------------------------------------------------ Keep Your Developer Skills Current with LearnDevNow! The most comprehensive online learning library for Microsoft developers is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3, Metro Style Apps, more. Free future releases when you subscribe now! http://p.sf.net/sfu/learndevnow-d2d _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
