| |
Primal zero-norm based feature selection by O.Mangasarian
|
A=FSV(H) returns a fsv object initialized with hyperparameters H.
Hyperparameters, and their defaults
lambda=1 -- the lambda parameter beteen 0 and 1.
(1=min. zero norm, 0=min. hinge loss)
alpha=5 -- the alpha parameter
feat=[] -- number of desired features (if empty gives
` the smallest)
output_rank=0 -- when set to 1, output features only
Model
rank -- the rank of the features
w -- the weight
b0 -- the threshold
Methods:
training, testing
Example:
d=gen(toy); a=fsv; a.feat=10; a.output_rank=1;[r,a]=train(a,d);
a.rank - lists the chosen features in order of importance
| |
Reference : Feature selection via concave minimization and support vector machines |
| |
Author : Paul S. Bradley and Olvi L. Mangasarian |