| |
Kernel Perceptron with optional margin.
|
A=DUALPERCEPTRON(H) returns a dualperceptron object initialized with hyperparameters H.
The dualperceptron object trains a potentially kernelized perceptron.
Hyperparameters (with defaults)
max_loops=100 -- Maximum number of sweeps through the data
margin=0 -- potential margin with which to train on
alpha_cutoff=-1; -- keep alphas with abs(a_i)>max(a)/alpha_cutoff
default keeps all alphas, another
reasonable choice is e.g alpha_cutoff=1e5 to remove
zero alphas (i.e non-SVs) to speed up computations.
Model
child=kernel -- the kernel is stored as a member called "child"
alpha -- the weights
Xsv -- the "Support Vectors"
Methods:
train, test
Example:
d=gen(toy2d);
[r,a]=train(dualperceptron('max_loops=20'),d);
plot(a)
d=gen(spiral({'m=200','n=2','noise=0.35'}));
[r,a]=train(dualperceptron(kernel('rbf',1)),d)
plot(a)
| |
Reference : Pattern Classification |
| |
Author : Richard O. Duda , Peter E. Hart |