| |
Probabilistic Principal Components Analysis
|
A=PPCA(H) returns a ppca object initialized with hyperparameters H.
Hyperparameters, and their defaults
feat=0; -- number of features, default 0 means all via dim X
iterations=10000 -- number of iterations for the EM algorithm
Model
e_val -- the eigenvalues
e_vec -- the eigenvectors
W -- the principal components
offset -- the mean of the training data
sigma -- the isotropic noise term sigma^2
dat -- training data (that we extracted from)
Methods:
train, test
Example:
d=gen(toy({'l=1000','n=2'}));
d.X=d.X*[1,0.1;0.2,0.2];
[r,a]=train(ppca('iterations=5000'),d);
plot(d.X(:,1),d.X(:,2),'r.');
hold on;
line([0,a.e_vec(1,1)],[0,a.e_vec(2,1)])
line([0,a.e_vec(1,2)],[0,a.e_vec(2,2)])
axis([-1,1,-1,1])
| |
Reference : Probabilistic analysis of kernel principal components : mixture modeling and classification |