GoDec的matlab代码

function [L,S,RMSE,error]=GoDec(X,rank,card,power) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % GoDec Algotithm %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %INPUTS: %X: nxp data matrix with n samples and p features %rank: rank(L)<=rank %card: card(S)<=card %power: >=0, power scheme modification, increasing it lead to better %accuracy and more time cost %OUTPUTS: %L:Low-rank part %S:Sparse part %RMSE: error %error: ||X-L-S||/||X|| %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %REFERENCE: %Tianyi Zhou and Dacheng Tao, "GoDec: Randomized Lo-rank & Sparse Matrix %Decomposition in Noisy Case", ICML 2011 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Tianyi Zhou, 2011, All rights reserved. %iteration parameters iter_max=1e+2; error_bound=1e-3; iter=1; RMSE=[]; %matrix size [m,n]=size(X); if m<n X=X'; end %initialization of L and S L=X; S=sparse(zeros(size(X))); tic; while true %Update of L Y2=randn(n,rank); for i=1:power+1 Y1=L*Y2; Y2=L'*Y1; end [Q,R]=qr(Y2,0); L_new=(L*Q)*Q'; %Update of S T=L-L_new+S; L=L_new; [Temp,idx]=sort(abs(T(:)),'descend'); S=zeros(size(X));S(idx(1:card))=T(idx(1:card)); %Error, stopping criteria T(idx(1:card))=0; RMSE=[RMSE norm(T(:))]; if RMSE(end)<error_bound || iter>iter_max break; else L=L+T; end iter=iter+1; end toc; LS=L+S; error=norm(LS(:)-X(:))/norm(X(:)); if m<n LS=LS'; L=L'; S=S'; end