knPcaPred is not finished. need test

chapter06/demo.m

@@ -47,8 +47,15 @@
 % maxabsdiff(sigma_kn,sigma_lin)
 % maxabsdiff(p_kn,p_lin)
 %% kernel PCA with linear kernel is PCA
-% clear; close all;
-% n = 100;
+clear; close all;
+d = 10;
+p = 2;
+n = 500;
+X = randn(d,n);
+
+model = pca(X,p);
+Y = pcaPred(model,X);
+
 %% test case for knCenter
 % kn = @knGauss;
 % X=rand(2,100);

chapter06/knPca.m

@@ -1,4 +1,4 @@
-function [ R, Z, err, model] = knPca(X, p, kn)
+function model = knPca(X, p, kn)
 % Kernel PCA
 %   X: dxn data matrix 
 %   p: target dimension
@@ -7,17 +7,16 @@
 if nargin < 3
     kn = @knGauss;
 end
-
 K = kn(X,X);
 K = knCenter(K);
-[V,A] = eig(K);
-[A,idx] = sort(diag(A),'descend');
-V = V(:,idx(1:p))';
-A = A(1:p);
-R = bsxfun(@times,V,1./sqrt(A));
+[V,L] = eig(K);
+[L,idx] = sort(diag(L),'descend');
+V = V(:,idx(1:p));
+L = L(1:p)';
+U = bsxfun(@times,V,1./sqrt(L));
 if nargout > 1    
-    Z = bsxfun(@times,V,sqrt(A));
+    Z = bsxfun(@times,V,sqrt(L));
 end
-if nargout > 2
-    err = diag(K)'-sum(Z.^2,1);
-end
+model.V = V;
+model.L = L;
+model.X = X;

chapter06/knPcaPred.m

@@ -1,9 +1,14 @@
-function X = knPcaPred(model, Xt)
+function X = knPcaPred(model, Xt, opt)
 % Prediction for kernel PCA
 %   model: trained model structure
 %   X: d x n testing data
 %   t (optional): 1 x n testing response
 % Written by Mo Chen ([email protected]).
 
+U = model.U;
+L = model.L;
 
+if nargin == 3 && opt.whiten
+    Y = ;
+end