It may have to do with the different scaling behaviors that these two types of penalties show with number of samples. I remember there being investigations into this for the respective sklearn classifiers, but I don't know the result, nor literature to back this up.
On Wed, Jul 22, 2015 at 3:58 PM, Doaa Altarawy <daltar...@vt.edu> wrote: > Thanks, that's looks the most suitable solution if I'll multiply the > weights in both L1 and L2. > > Currently the paper I use is following the Adaptive elastic net, that is > the weights are multiplied only in the Lasso penalty. I don't know what is > the effect if the weights are multiplied in the Ridge penalty too. > I revised the paper but they didn't say why it is multiplied in L1 only > but not in L2. > > Any ideas about that? > > -- > Doaa Altarawy, > PhD Student, Computer Science, > Virginia Tech, USA. > > > > > ------------------------------------------------------------------------------ > > _______________________________________________ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > >
------------------------------------------------------------------------------
_______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
