2012/2/27 Matthias Ekman <[email protected]>: > thanks for all the helpful remarks! That's exactly what I wanted to > know. However I am a bit surprised by the low performance of Elastic Net > in comparison to logit (both using L1 regularization and test/training > on the full dataset). Am I overseeing something obvious here? > > acc enet: 69.0 > acc logit: 100.0 > > iris = datasets.load_iris() > X = iris.data > y = iris.target > > idx=np.where(y!=2)[0] # select only labels 0 & 1 > X=X[idx,:] > y=y[idx,:] > X /= X.std(0) > > clf = OneVsRestClassifier(ElasticNet(alpha=0., rho=0.)) > y_pred = clf.fit(X,y).predict(X) > print 'acc enet:',zero_one_score(y,y_pred)*100
Why alpha and rho to 0? Usual rho is good around 0.8 and alpha should be adjusted by grid search. -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel ------------------------------------------------------------------------------ Try before you buy = See our experts in action! 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-dev2 _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
