diff util/classes/dictionaryMatrices/rotatematrix.m @ 170:68fb71aa5339 danieleb

Added dictionary decorrelation functions and test script for Letters paper.

--- a/util/classes/dictionaryMatrices/rotatematrix.m	Thu Sep 29 09:46:52 2011 +0100
+++ b/util/classes/dictionaryMatrices/rotatematrix.m	Thu Oct 06 14:33:41 2011 +0100
@@ -5,54 +5,66 @@
 % REFERENCE
 % M.D. Plumbley, Geometrical Methods for Non-Negative ICA: Manifolds, Lie
 % Groups and Toral Subalgebra, Neurocomputing
+
+%% Parse inputs and set defaults
 if ~nargin, testrotatematrix; return, end
 
+if ~exist('param','var')  || isempty(param),  param  = struct; end
+if ~exist('method','var') || isempty(method), method = 'conjgradLie'; end
+if ~isfield(param,'nIter'), param.nIter = 100; end	%number of iterations
+if ~isfield(param,'eps'), param.eps = 1e-9; end		%tolerance level
+if ~isfield(param,'step'), param.step = 0.01; end
 
-if ~exist('method','var') || isempty(method), method = 'unconstrained'; end
+J = @(W) 0.5*norm(D-W*Phi,'fro');		%cost function
 
-J = @(W) 0.5*norm(D-W*Phi,'fro');
-cost = zeros(param.nIter,1);
+% Initialise variables
+cost = zeros(param.nIter,1);			%cost at each iteration
+W	 = eye(size(Phi,1));				%rotation matrix
+grad = ones(size(W));					%gradient
+t	 = param.step;						%step size
+Gprev = 0;								%previous gradient
+Hprev = 0;								%previous Lie search direction
+iIter = 1;								%iteration counter
 
-W   = eye(size(Phi,1));
-t = 0;
-Gprev = 0;
-Hprev = 0;
-for i=1:param.nIter
-	cost(i) = J(W);
-	grad = (W*Phi-D)*Phi';
+%% Main algorithm
+while iIter<=param.nIter && norm(grad,'fro')>eps
+	cost(iIter) = J(W);				%calculate cost
+	grad = (W*Phi-D)*Phi';			%calculate gradient
 	switch method
 		case 'unconstrained'		% gradient descent
 			eta = param.step;
 			W = W - eta*grad;		% update W by steepest descent
 		case 'tangent'				% self correcting tangent
 			eta = param.step;
-			mu  = param.reg;
-			W = W - 0.5*eta*(grad - W*grad'*W + mu*W*(W'*W-eye(size(W))));
-		case 'steepestlie'
+			W = W - 0.5*eta*(grad - W*grad'*W);
+			[U , ~, V] = svd(W);
+			W = U*V';
+		case 'steepestlie'			%steepest descent in Lie algebra
 			eta = param.step;
-			B = 2*skew(grad*W');	% calculate gradient in lie algebra
+			B = 2*skew(grad*W');	% calculate gradient in Lie algebra
 			W = expm(-eta*B)*W;		% update W by steepest descent
-		case 'linesearchlie'
-			B = 2*skew(grad*W');	% calculate gradient in lie algebra
+		case 'linesearchlie'		% line search in Lie algebra
+			B = 2*skew(grad*W');	% calculate gradient in Lie algebra
 			H = -B;					% calculate direction as negative gradient
-			t = searchline(J,H,W,t);% line search in one-parameter lie subalgebra
+			t = searchline(J,H,W,t);% line search in one-parameter Lie subalgebra
 			W = expm(t*H)*W;		% update W by line search
-		case 'conjgradlie'
-			G = 2*skew(grad*W');	% calculate gradient in lie algebra
+		case 'conjgradlie'			% conjugate gradient in Lie algebra
+			G = 2*skew(grad*W');	% calculate gradient in Lie algebra
 			H = -G + polakRibiere(G,Gprev)*Hprev; %calculate conjugate gradient direction
-			t = searchline(J,H,W,t);% line search in one-parameter lie subalgebra
+			t = searchline(J,H,W,t);% line search in one-parameter Lie subalgebra
 			W = expm(t*H)*W;		% update W by line search
-			Hprev = H;				% % save search direction
-			Gprev = G;				% % save gradient
+			Hprev = H;				% save search direction
+			Gprev = G;				% save gradient
 	end
+	iIter = iIter+1;				% update iteration counter
 end
-Dhat = W*Phi;
+Dhat = W*Phi;						%rotate matrix
+cost(iIter:end) = cost(iIter-1);	%pad cost vector
 end
-% function C = matcomm(A,B)
-% %Matrix commutator
-% C = A*B-B*A;
 
+%% Support functions
 function gamma = polakRibiere(G,Gprev)
+%Polak-Ribiere rule for conjugate direction calculation
 gamma = G(:)'*(G(:)-Gprev(:))/(norm(Gprev(:))^2);
 if isnan(gamma) || isinf(gamma)
 	gamma = 0;
@@ -60,36 +72,41 @@
 end
 
 function t = searchline(J,H,W,t)
+%Line search in one-parameter Lie subalgebra
 t = fminsearch(@(x) J(expm(x*H)*W),t);
 end
 
 function B = skew(A)
+%Skew-symmetric matrix
 B = 0.5*(A - A');
 end
 
 
+%% Test function
 function testrotatematrix
 clear, clc, close all
-n = 256;
-m = 512;
-disp('A random matrix...');
-Phi = randn(n,m);
-disp('And its rotated mate...');
-Qtrue = expm(skew(randn(n)));
-D = Qtrue*Phi; 
-disp('Now, lets try to find the right rotation...');
-param.nIter = 1000;
-param.step  = 0.001;
+n = 256;							%ambient dimension
+m = 512;							%number of atoms
+param.nIter = 300;				%number of iterations
+param.step  = 0.001;			%step size
+param.mu    = 0.01;			%regularization factor (for tangent method)
+methods = {'unconstrained','tangent','linesearchlie','conjgradlie'};
 
-cost = zeros(param.nIter,4);
-[~, cost(:,1)] = rotatematrix(D,Phi,'unconstrained',param);
-[~, cost(:,2)] = rotatematrix(D,Phi,'steepestlie',param);
-[~, cost(:,3)] = rotatematrix(D,Phi,'linesearchlie',param);
-[~, cost(:,4)] = rotatematrix(D,Phi,'conjgradlie',param);
+Phi = randn(n,m);				%initial dictionary
+Qtrue = expm(skew(randn(n)));	%rotation matrix
+D = Qtrue*Phi;					%target dictionary
+
+cost = zeros(param.nIter,length(methods));
+for iIter=1:length(methods)
+	tic
+	[~, cost(:,iIter)] = rotatematrix(D,Phi,methods{iIter},param);
+	time = toc;
+	sprintf('Method %s completed in %f seconds \n',methods{iIter},time)
+end
 
 figure, plot(cost)
 set(gca,'XScale','log','Yscale','log')
-legend({'uncons','settpestlie','linesearchlie','conjgradlie'})
+legend(methods)
 grid on
 xlabel('number of iterations')
 ylabel('J(W)')