covfixer2.m

关于高斯混合模型（GMM）的matlab源代码：

 function [nsigma, varargout] = covfixer2(sigma);
 %COVFIXER2   - force matrix to be a valid covariance matrix
 %
 % covmatrix = COVFIXER2(matrix)
 % Matrix is forced (complex conjugate) symmetric,
 % positive definite and its diagonal real valued.
 %
 % [covmatrix, loops] = COVFIXER2(...)
 %    loops - number of rounds the positive definite fixer had to run.
 %
 % [covmatrix, loops, symerr] = COVFIXER2(...)
 %    symerr - symmetry error matrix
 
 %
 % $Name:  $
 % $Id: covfixer2.m,v 1.2 2004/08/18 07:52:12 paalanen Exp $
 % Copyright 2003, Pekka Paalanen <[e-mail]>
 
 % except isspd() function which is from The MathWorks Matlab mvnpdf.m.
 
 D = size(sigma, 1);
 fixrate = 0.01;
 covfixmat = ones(D) + fixrate*eye(D);
 loops = 0;
 min_limit = eps*10;
 
 if ~all( isfinite(sigma(:))  )
     error('covariance matrix is not finite');
 end
 
 % Running imagfixer is not counted as covariance fixing,
 % the changes are assumed to be so small.
 nsigma = imagfixer(sigma);
 
 if nargout>2
     varargout(2) = {(sigma-nsigma)};
 end
 
 while isspd(nsigma) == 0
     % covariance matrix is not positive definite
     % fix it
     loops  = loops+1;
     d = diag(nsigma);
     if any(d <= min_limit)
         % negative or zero (<eps) on the diagonal
         m = max(abs(d)) * fixrate;
         neg = min(d);
         if neg < 0
             % there is a negative component on the diagonal
             % get rid of it.
             addit = (m-neg)*eye(D);
         else
             if m < min_limit
                 m = min_limit;
             end
             addit = m*eye(D);
         end
         nsigma = nsigma + addit;
     else
         % increase diagonal values by 1 percent
         nsigma = nsigma .* covfixmat;
     end
 end
 
 if nargout>1
     varargout(1) = {loops};
 end
 
 
 % ------------------
 
 function [t,R] = isspd(Sigma)
 %ISPDS Test if a matrix is positive definite symmetric
 % T = ISPDS(SIGMA) returns a logical indicating whether the matrix SIGMA is
 % square, symmetric, and positive definite, i.e., it is a valid full rank
 % covariance matrix.
 %
 % [T,R] = ISPDS(SIGMA) returns the cholesky factor of SIGMA in R.  If SIGMA
 % is not square symmetric, ISPDS returns [] in R.
 
 %   Copyright 1993-2002 The MathWorks, Inc.
 %   Revision: 1.2   Date: 2002/03/28 16:51:27
 
 % Test for square, symmetric
 % NOTE: imagfixer already enforces squareness and symmetricity,
 % and fixing affects only the diagonal, so this is not necessary
 %[n,m] = size(Sigma);
 %if (n == m) & all(all(abs(Sigma - Sigma') < 10*eps*max(abs(diag(Sigma)))));
 
     % Test for positive definiteness
     [R,p] = chol(Sigma);
     if p == 0
         t = 1;
     else
        t = 0;
    end
 
 %else
 %    R = [];
 %    t = 0;
 %end
 
 % ------------------
 
 function nsigma = imagfixer(sigma);
 
 % force symmetric
 nsigma = sigma - (sigma - sigma')/2;
 % purge imag
 purge = imag(diag(nsigma));
 nsigma = nsigma - diag(purge)*1i;
 
 if max(purge) > 1e-4
     warning_wrap('gmmbayes:covfixer2:imagfixer', 'Quite big imaginary components removed from the diagonal');
 end