On Sat, Mar 17, 2012 at 4:44 AM, Alexandre Gramfort <[email protected]> wrote: > without the scale_C the libsvm/liblinear bindings are the only models > whose hyperparameters > depend on the training set size.
This statement doesn't sound true. Generally hyper-parameters (especially ones to do with regularization) *do* depend on training set size, and not in such straightforward ways. Data is never perfectly I.I.D. and sometimes it can be far from it. My impression was that standard practice for SVMs is to optimize C on held-out data. When would the scale_C heuristic actually save anyone from having to do this optimization? Even if the scale_C heuristic (is it fair to call it that?) is a good idea, My 2c is that it does not justify redefining the meaning of the "C" parameter which has a very standard interpretation in papers, textbooks, and other SVM solvers. If you really must redefine the C parameter (but why?) then it would make sense to me to rename it as well. - James ------------------------------------------------------------------------------ This SF email is sponsosred by: Try Windows Azure free for 90 days Click Here http://p.sf.net/sfu/sfd2d-msazure _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
