Mercurial > hg > smallbox
diff DL/Majorization Minimization DL/ExactDicoRecovery/mmdl_cn.m @ 155:b14209313ba4 ivand_dev
Integration of Majorization Minimisation Dictionary Learning
author | Ivan Damnjanovic lnx <ivan.damnjanovic@eecs.qmul.ac.uk> |
---|---|
date | Mon, 22 Aug 2011 11:46:35 +0100 |
parents | |
children |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/DL/Majorization Minimization DL/ExactDicoRecovery/mmdl_cn.m Mon Aug 22 11:46:35 2011 +0100 @@ -0,0 +1,26 @@ +% Majorization Minimization for Dictionary Learning +% Y = input data (M X L matrix) +% Phi = initial dictionary (M X N), e.g. random dictionary or first N data samples +% lambda = regularization coefficient (||Phi*X-Y||_F)^2 + lambda*||X||_1 +% IT = number of iterations +% res = dictionary constraint. 'un' = unit colomn norm, 'bn' = bounded colomn norm +function [Phiout,X,ert] = mmdl_cn(Y,Phi,lambda,IT,res) +maxIT = 1000; +[PhiN,PhiM] = size(Phi); +RR1 = PhiM; +%%%%%%%%%%%%%% +[PhiN,L] = size(Y); +X = ones(PhiM,L); +for it = 1:IT + to = .1+svds(Phi,1); + [PhiN,PhiM] = size(Phi); + %%%% + eps = 3*10^-4; + map = 1; % Projecting on the selected space (0=no,1=yes) + [X,l1err] = mm1(Phi,Y,X,to,lambda,maxIT,eps,map); %% Sparse approximation with Iterative Soft-thresholding + ert(it) = l1err; + %%% + eps = 10^-7; + [Phi,X] = dict_update_REG_cn(Phi,Y,X,maxIT,eps,res); +end +Phiout = Phi; \ No newline at end of file