Notes on the preliminary results of a linear two-class classifier in the PERMON toolbox
Marek Pecha, Václav Hapla, David Horák, Martin Čermák · AIP conference proceedings · 2018
This paper deals with the preliminary results of a linear two-class Support Vector Machines (SVM) implemented in PERMON toolbox. We present the first insights into training PermonSVM classifier using quadratic programming (QP) algorithms from the PemonQP, i.e. Dostál’s SMALBE, which is based on the augmented Lagrangian approach, and MPGP algorithms for box constrained QP. In presented benchmark on the URL dataset, we analyze the abilities of the QP solver with the respect to regularized parameter C and QP solver accuracy eps. In fact, we consider eps as the second parameter of the linear SVM and therefore we got better information about the tested algorithm behaviour.