Linear SVM with general regularization 1.0

remi flamary,, alain rakotomamonjy — Fri, 05 Oct 2012 15:34:21 -0000

We provide a general solver for squared hinge loss SVM of the form:

min_{w,b} sum_i max(0,y_i(x_i^Tw+b))^2 + Omega(w)

where Omega(w) can be :

l1 : Omega(w)=sum_i |w_i|
l2 : Omega(w)=sum_i |w_i|^2
l1-l2: Omega(w)=sum_g ||w_g||_2
l1-lp: Omega(w)=sum_g ||w_g||_p
adaptive l1-l2: Omega(w)=sum_g beta_g||w_g||_2

We also provide a multitask solver where T tasks can be learned simultaneously with joint sparsity constraints (mixed norm regularization).

Note that this toolbox has been designed to be efficient for dense data whereas most of the existing linear svm solvers have been designed for sparse datasets.

mloss.org Linear SVM with general regularization

Linear SVM with general regularization 1.0