it could also be that folds are different and that LIBSVM does maybe some per fold scaling. I would look at the scores on each grid point to see if the output of libsvm gives a similar score.
Alex On Tue, Aug 14, 2012 at 2:18 PM, Andreas Müller <[email protected]> wrote: > Hi Jose. > We are using the LIBSVM library as backend, so in principle the results > should be identical. It could be that the default tolerance (tol) is > different. > > Which version of sklearn are you using? > > Cheers, > Andy > > ----- Ursprüngliche Mail ----- > Von: "José Guilherme Camargo de Souza" <[email protected]> > An: [email protected] > Gesendet: Dienstag, 14. August 2012 13:16:19 > Betreff: [Scikit-learn-general] Differences between pure LIBSVM and > Scikit-learn SVR implementation > > Hi all, > > I'm new to scikit-learn and I'm trying to reproduce some results > obtained with LIBSVM. I'm using the SVM regression and doing grid > search to find the optimal C, gamma and epsilon for my training > dataset. > > The ranges for the search using libsvm implementation are: > > C = [1, 10] > gamma = [0.0001, 0.01] > epsilon= [0.1, 0.2] > > When I try the same values with the grid search + SVR implemented > in Scikit-learn I get different results. I wonder how should I adapt > the values above to the scikit-learn implementation. Does anyone have > some pointers? > > Thanks, > > > José Guilherme > > ------------------------------------------------------------------------------ > Live Security Virtual Conference > Exclusive live event will cover all the ways today's security and > threat landscape has changed and how IT managers can respond. Discussions > will include endpoint security, mobile security and the latest in malware > threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ > _______________________________________________ > Scikit-learn-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > ------------------------------------------------------------------------------ > Live Security Virtual Conference > Exclusive live event will cover all the ways today's security and > threat landscape has changed and how IT managers can respond. Discussions > will include endpoint security, mobile security and the latest in malware > threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ > _______________________________________________ > Scikit-learn-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
