This program minimizes f, the radius margin bound of LOO error for the SVM formulation in which quadratic penalties are used for margin plane violations. f, which is a function of c and sigmasq, is evaluated using efficient nearest point/SMO algorithms. ---------------------------------------------------------------- List of c, sigmasq, the corresponding f & num_of_sup_vectors obtained during the optimization process: ---------------------------------------------------------------- 0.10000000D+01 0.20000000D+01 0.8718783 378 0.70745539D+00 0.78265948D+00 0.5600647 302 0.51852918D+00 0.48452415D+00 0.4911393 299 0.38863966D+00 0.31205688D+00 0.4630062 324 0.35101362D+00 0.23729592D+00 0.4577116 332 0.35940953D+00 0.21493326D+00 0.4566522 333 0.38990722D+00 0.19490121D+00 0.4558111 327 0.40277721D+00 0.19354077D+00 0.4557383 326 0.40422789D+00 0.19445437D+00 0.4557361 325 0.40431502D+00 0.19470456D+00 0.4557360 325 ---------------------------------------------------------------- Total number of evaluations of f: 10 Final c, sigmasq: 0.40431502D+00 0.19470456D+00 Value of threshold, b: 0.70210987D-01 Number of support vectors: 325 ---------------------------------------------------------------- Details of support vectors and their associated Lagrangian multipliers (alphas) are printed in a separate user-specified output file. ---------------------------------------------------------------- Num of MisClassifications on Test Set: 561 Percentage MisClassifications on Test Set: 11.44898 ----------------------------------------------------------------