pca and pcaPred are rewritten (not finished).

chapter06/pca.m

@@ -0,0 +1,13 @@
+function model = pca( X, p )
+% Principal component analysis
+% Written by Mo Chen ([email protected]).
+
+Xo = bsxfun(@minus,X,mean(X,2));
+S = Xo*Xo'/size(X,2);                   % 12.3
+[U,A] = eig(S);                         % 12.5
+[A,idx] = sort(diag(A),'descend');      
+U = U(:,idx(1:p));
+A = A(1:p);
+
+model.U = U;
+model.A = A;

chapter06/pcaPred.m

@@ -0,0 +1,6 @@
+function  = pcaPred( model, Xt )
+% Prediction for PCA: project future data to principal subspace
+%   model: trained model structure
+%   Xt: d x n testing data
+% Written by Mo Chen ([email protected]).
+

chapter06/pcaPred.m~

@@ -0,0 +1,6 @@
+function  = pcaPred( model, Xt )
+% Prediction for PCA: project future data to principal subspace
+%   model: trained model structure
+%   Xt: d x n testing data
+% Written by Mo Chen ([email protected]).
+

chapter12/pca.m~

@@ -0,0 +1,40 @@
+function [V, A] = pca(X, p)
+% Perform standard PCA (spectral method).
+%   X: d x n data matrix
+%   p: dimension of target space (p>=1) or ratio (0<p<1)
+% Written by Mo Chen ([email protected]).
+
+Xo = bsxfun(@minus,X,mean(X,2));
+
+
+opts.disp = 0;
+opts.issym = 1;
+opts.isreal = 1;
+opts.maxit = 500;
+
+[d,n] = size(X);
+if nargin == 1
+    p = min(d,n);
+end
+
+
+if 0<p && p<1       % given ratio
+    [V,A] = svd(X,'econ');
+    A = diag(A).^2;
+
+    S = cumsum(A);
+    idc = (S/S(end))<=p;
+    V = V(:,idc);
+    A = A(idc);
+elseif p >= min(d,n) % full pca
+    [V,A] = svd(X,'econ');
+    A = diag(A).^2;
+elseif d <= n       % covariance based pca
+    [V,A] = eigs(X*X',p,'la',opts); 
+    A = diag(A);
+elseif d > n                % inner product based pca
+    [U,A] = eigs(X'*X,p,'la',opts); 
+    A = diag(A);
+    V = X*bsxfun(@times,U,1./sqrt(A)');
+end
+