RSPOLY2

Reduced set method for second order homogeneous polynomial kernel.



 Synopsis:
  red_model = rspoly2(model)
  red_model = rspoly2(model,max_nsv)

 Description:
  It uses reduced set techique to reduce complexity
  of the kernel expansion with second order homogeneous polynomial 
  kernel k(x,y) = (x'*y)^2 = kernel(x,y,'poly',2) .

  The method was published in 
  J.C.Burges: Simplified Support Vector Decision Rules. ICML, 1996.
  
 Input:
  model [struct] Kernel expansion:
   .Alpha [nsv x 1] Weights of kernel expansion.
   .b [1x1] Bias.
   .sv.X [dim x nsv] Support vectors.
   .options.ker = 'poly'
   .options.arg = [2 0]

  max_nsv [1x1] Maximal number of new support vectors. If not given 
   then the new expansion approximates the original one exactly with
   at most dim support vectors. 

 Output:
  red_model [struct] Reduced kernel expansion:
  red_model.Alpha [new_nsv x 1] New weights.
  red_model.b [scalar] Bias.
  red_model.sv.X [dim x new_nsv] New support vectors.
  ...

 Example:
  trn = load('riply_trn');
  model = smo(trn,struct('ker','poly','arg',[2 0],'C',10));
  red_model = rspoly2( model );
  figure; 
  subplot(1,2,1); axis square; ppatterns(trn); psvm(model);
  subplot(1,2,2); axis square; ppatterns(trn); psvm(red_model);
  
 See also
  RSRBF.


Source: rspoly2.m

About: Statistical Pattern Recognition Toolbox
(C) 1999-2004, Written by Vojtech Franc and Vaclav Hlavac
Czech Technical University Prague
Faculty of Electrical Engineering
Center for Machine Perception

Modifications:
22-dec-2004, VF, header and comments added
28-nov-2003, VF