REDQUADH |
reduced SVM classifier with homogeneous quadratic kernel.
Synopsis:
red_model = redquadh(model)
Description:
It uses reduced set techique (Burges) to compute
simpler SVM binary rule with homogeneous quadratic kernel (x'*y)^2.
Input:
model.Alpha [nsv x 1] Weights of kernel expansion.
model.b [scalar] Bias.
model.sv.X [dim x nsv] Support vectors.
model.options.ker = 'poly'
model.options.arg = [2 0]
Output:
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,{'ker','poly','arg',[2 0],'C',10});
red_model = redquadh( model );
figure; ppatterns(trn); psvm(model);
figure; ppatterns(trn); psvm(red_model);
Modifications:
28-nov-2003, VF