mogp.m

数据挖掘的工具箱,最新版的,希望对做这方面研究的人有用

function p = mogP(x,means,covs,priors)
%MOGP Compute the probability density of a Mixture of Gaussians
%
%       P = MOGP(X,MEANS,COVS,PRIORS)
%
% Calculate the probability density for all objects X for all of the
% clusters of a mixture of Gaussians, characterized with the MEANS,
% COVS and PRIORS.
% Note that P is not normalized!
%
% See also: mogEM, mog_dd

% Copyright: D.M.J. Tax, R.P.W. Duin, [e-mail]
% Faculty of Applied Physics, Delft University of Technology
% P.O. Box 5046, 2600 GA Delft, The Netherlands

% Get the useful parameters
[N,d] = size(x);
k = length(priors);
p = zeros(N,k);

% First detect which type of covariance matrix we have:
covtype = ndims(covs);
if ((covtype==2)&(size(covs,2)==1))
	covtype = 1;
end
% Depending on the covariance matrix, the p is computed differently:
switch covtype
case 1 % Diagonal cov.matrix with equal variances
	D = distm(x,means);
	sig = 2.*repmat(covs',N,1);
	Z = (pi*sig).^(d/2);
	p = exp(-(D./sig))./Z;
case 2 % Diagonal cov.matrix with unequal variances
	Z = (2*pi).^(d/2);
	sig = prod(sqrt(covs),2);
	for i=1:k
		dif = x - repmat(means(i,:),N,1);
		p(:,i) = exp(-sum((dif.*dif)./(repmat(covs(i,:),N,1)) ,2)/2) ./ ...
		                                  (Z*sig(i));
	end
case 3 % Complete covariance matrix
	Z = (2*pi).^(d/2);
	for i=1:k
		dif = x - repmat(means(i,:),N,1);
		c = chol(squeeze(covs(i,:,:)));
		Dmah = dif/c; % here is the obfuscated inverse...
		p(:,i) = exp(-sum(Dmah.*Dmah,2)/2) ./ (Z*prod(diag(c)));
	end
otherwise
	error('The covariance matrix parameter is not well-defined');
end

% include the priors:
p = p.*repmat(priors,N,1);

return