Florian Hönig <florian.hoenig@...> writes: > Dear list, > > analogously to sklearn.preprocessing.scale and sklearn.preprocessing.Scaler, > I would like to add something for scaling the individual features to > the interval [0;1]. > > I have encountered a number of datasets where mean/variance scaling didn't > help > much for SVM/SVR, while scaling to [0;1] worked miraculously.
Florian, folks, normalizing all |x| to 1, which amounts to using cosine distance, is good for sparse data, *sometimes*; see http://stats.stackexchange.com/questions/29627/euclidean-distance-is-usually-not-good-for-sparse-data But I think scaling should be left to users -- even 3 or 4 sizes cannot fit all data. Florian, is your data images, could you say "scale to [0:1] when ..." ? Thanks, cheers -- denis ------------------------------------------------------------------------------ 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
