A new algorithm for computation of a regularization solution path for reinforced multicategory support vector machines

The recently proposed Reinforced Multicategory Support Vector Machine (RMSVM) has been proven to have desirable theoretical properties as well as competitive numerical accuracy for multi-class classification problems. Currently solving the RMSVM is based on a grid search approach for selecting the tuning parameter , which dramatically increases its computational complexity. To overcome this hurdle we develop a new algorithm RMSVMPATH to compute a regularization solution path for RMSVM. We relax the commonly used continuity assumption and propose a new linear programming approach. Numerical simulations and real data analyses demonstrate that the proposed algorithm can yield a valid solution path at a low computational cost. The Canadian Journal of Statistics 45: 149-163; 2017 (c) 2017 Statistical Society of Canada