MATLAB: SIGMA must be a square, symmetric, positive definite matrix

bayesianclassifierpositivesigmasquaresymmetric

Hello, I've a problem with the function mvnpdf. The error is: SIGMA must be a square, symmetric, positive definite matrix. But for me SIGMA is square, symmetric and positive. This is my code for bayesian classifier:
m=size(T{1},2);
M=zeros(K,m);
Sig=zeros(m,m,K);
n=zeros(1,K);
for k=1:K
    X=T{k};
    n(k)=size(X,1);
    M(k,:)= mean(X);
    Sig(:,:,k)=cov(X);
end
p=n/sum(n); 
LV=[];
LC=[];
     m=length(S);
       for j=1:m
            LV=[LV;k];
            x=S(j,:);
            pdf=[];
        for l=1:K
              pdf=[pdf mvnpdf(x,M(l,:),Sig(:,:,l))];
        end
       P=p.*pdf;
       [v e]=max(P);
       LC=[LC; e];
      end
The values and dataset are in memory. Can someone please help me? Thanks.

Best Answer

A matrix of all NaN values (page 4 in your array) is most certainly NOT positive definite.
Related Question