Yashodhan Athavale

I tried modifying the code a little bit, by including the following lines of code,

function [W,M,V] = Init_EM(X,k)
[n,d] = size(X);
[Ci,C] = kmeans(X,k,'Start','cluster', ...
    'Maxiter',100, ...
    'EmptyAction','drop', ...
    'Display','off'); % Ci(nx1) - cluster indeices; C(k,d) - cluster centroid (i.e. mean)
while sum(isnan(C))>0,
    [Ci,C] = kmeans(X,k,'Start','cluster', ...
        'Maxiter',100, ...
        'EmptyAction','drop', ...
        'Display','off');
end
M = C';

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Use this routine if the cluster matrix is of variable length and contains
% Scan the cluster index matrix Ci for NaN values, and remove them
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
temp1=Ci(:,1); % Scan the entire matrix column
temp2=isnan(temp1); % Find the rows with NaN values
index_active=find(temp2~=1); % Find the rows with actual values
Ci_new=temp1(index_active); %#ok<FNDSB>
[r,c]=size(Ci_new); % Determine the size of the newly scanned matrix.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Vp = repmat(struct('count',0,'X',zeros(r,c)),1,k);
for i=1:r % Separate cluster points
    Vp(Ci_new(i)).count = Vp(Ci_new(i)).count + 1;
    Vp(Ci_new(i)).X(Vp(Ci_new(i)).count,:) = X(i,:);
end

V = zeros(d,d,k);
for i=1:k,
    W (i) = Vp(i).count/r;
    % Find the covariance ignoring the NaNs
    V(:,:,i) = nancov(Vp(i).X(1:Vp(i).count,:));
end
%%%%%%%%%%%%%%%%%%%%%%%%
%%%% End of Init_EM %%%%
%%%%%%%%%%%%%%%%%%%%%%%%

I worked, giving me the estimates. But still I am getting some warnings with "Empty rows of data".