GERMAN 
------

results of the model selection (cf. [RaeOnoMue01,MikRaeWesSchMue99])
---------------------------------------------------
  RBF-Network               : 8 rbf-centers; 3 steps of rbf-opt.
  AdaBoost with RBF-Network : 200 boost.iter.; 8 centers; 3 steps of rbf-opt.
  LP_Reg-AdaBoost (-"-)     : C=5.000e-06; 200 boost.iter.; 8 centers; 3 steps of rbf-opt.
  QP_Reg-AdaBoost (-"-)     : C=1.000e-04; 200 boost.iter.; 8 centers; 3 steps of rbf-opt.
  AdaBoost_Reg (-"-)        : C=1.500e+04; 200 boost.iter.; 8 centers; 3 steps of rbf-opt.
  SVM with RBF-Kernel       : C=3.162e+00; sigma=5.500e+01
  KFD with RBF-Kernel       : C=10; sigma=35; mu=1

averaged classification error on the test set in % (cf. [RaeOnoMue01,MikRaeWesSchMue99])
-----------------------------------------------------------------------
  RBF-Network               : 24.71 +/- 2.38
  AdaBoost with RBF-Network : 27.45 +/- 2.50
  LP_Reg-AdaBoost (-"-)     : 24.79 +/- 2.22
  QP_Reg-AdaBoost (-"-)     : 25.25 +/- 2.14
  AdaBoost_Reg (-"-)        : 24.34 +/- 2.08
  SVM with RBF-Kernel       : 23.61 +/- 2.07
  Kernel Fisher Discriminant: 23.71 +/- 2.20

  cf. with the list in http://ida.first.gmd.de/~raetsch/data/german/german.errors,
  which contains the classification error rates of the classifiers
  for each realization of the data.

references:
-----------
@article{RaeOnoMue01,
  author =	 {G.~R{\"a}tsch and T.~Onoda and K.--R.~M{\"u}ller},
  title =	 "Soft Margins for {A}da{B}oost",
  year =	 2001,
  journal =	 "Machine Learning",
  note =	 "also NeuroCOLT Technical Report NC-TR-1998-021",
  publisher =	 "Kluwer Academic Publishers",
  month =	 mar,
  volume =	 42,
  number =	 3,
  pages =	 "287--320",
  annote =       {http://www.boosting.org/papers/RaeOnoMue00a.pdf}
}

@inproceedings{MikRaeWesSchMue99,
  author =	 {S.~Mika and G.~R{\"a}tsch and J.~Weston and
                  B.~Sch{\"o}lkopf and K.--R.~M{\"u}ller},
  title =	 {Fisher Discriminant Analysis with Kernels},
  booktitle =	 {Neural Networks for Signal Processing IX},
  year =	 {1999},
  editor =	 {Y.-H.~Hu and J.~Larsen and E.~Wilson and S.~Douglas},
  publisher =	 {IEEE},
  pages =	 {41--48},
  annote =       {http://ida.first.gmd.de/~mika/PS/MikRaeWesSchMue99.pdf}
}

see also:

@PhdThesis{Rae01,
  author =	 {G.~R{\"a}tsch},
  title =	 {Robust Boosting via Convex Optimization},
  school =	 {University of Potsdam},
  year =	 2001,
  month =	 {October},
  address =	 {Neues Palais 10, 14469 Potsdam, Germany},
  annote =	 {http://www.boosting.org/papers/thesis.pdf}
}