Hi everybody. At the moment I'm trying to understand feature selection. I was looking at the "L1 based feature selection" that is described in the docs.
I was trying to use that with LinearSVC but I don't really understand what is going on. Maybe someone can explain. I am in the mult-class setup. There, the CoefSelectTransformerMixin takes the mean over the different classes and thresholds. This seems quite weird to me. Taking the mean means that if a feature has a strong positive weight for one class and a strong negative weight for another class, they might cancel, leading to the feature being not present in the solution. Why does that make sense? Or should the absolute value be taken before? The other thing is the thresholding. The thresholding is also done without taking the absolute value. This means all features that are negative on average (whatever that means?) are left out. Any help would be appreciated. Btw, there is no docstring in the "transform" method of CoefSelectTransformerMixin, meaning there is not docstring for the transform methods of the classes using it (i.e. LinearSVC). If someone knows how this mixin works, maybe they might be so kind as to write one ;) Cheers, Andy ------------------------------------------------------------------------------ Keep Your Developer Skills Current with LearnDevNow! 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-d2d _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
