| |
Semi-Supervised Learning by Zhou et al.
|
a=lgcz(hyperParam)
Generates a lgcz object with given hyperparameters.
Can use unlabeled data (Y=0) to solve classification problem.
After label propagation a SVM is trained to provide an inductive model.
Hyperparameters (with defaults)
child=kernel -- the kernel is stored as a member called "child"
propkern=kernel -- the kernel which is used for propagation
ridge=1e-13 -- a ridge on the kernel
wide -- the width of label propagation
Model
alpha -- lagrangian multipliers
Xsv -- support vectors
Methods:
train, test, get_w
Example:
c1=[2,0];
c2=[-2,0];
X1=randn(50,2)+repmat(c1,50,1);
X2=randn(50,2)+repmat(c2,50,1);
Y= 0*[ones(50,1);-ones(50,1)]; kill all labels
Y(1)=1;
Y(end)=-1; provide only two labelled points
d=data([X1;X2],Y);
clf;
hold on;
l=lgcz('ridge=1e-10');
l.child=kernel('rbf',1)
l.propkernel=kernel('rbf',0.1)
[r,a]=train(l,d)
plot(a)
p=plot(d.X(1,1),d.X(1,2),'go');set(p,'MarkerFaceColor',[0,1,0]);
p=plot(d.X(end,1),d.X(end,2),'go');set(p,'MarkerFaceColor',[0,1,0]);
['only green dots were labelled']
| |
Reference : Learning with Local and Global Consistency |