Vectorization of Weighted Minkowski Distance

I am trying to vectorize the code for this Weighted Minkowski Distance.

I am aware that in Matlab both pdist and pdist2 suffice the purpose when Theta is equal to either 1 or scalar:

Mat = theta*(squareform(pdist(X,'minkowski',p))) %for theta = scalar

However in my implementation, which follows, Theta is a vector with the same dimension as the number of columns of X.

X = [1:5; 2:6; 3:7]';
p=1.99;
n = length(X);
theta = [0.1 .7 .2];
tic
Mat = zeros(n,n);
for i=1:n
    for j=i+1:n
        Mat(i,j)=sum(theta.*abs(X(i,:)-X(j,:)).^p)^(1/p);
    end
end
Mat = Mat+Mat'+eye(n);
toc

In order to vectorize this, I made the following attempt, but I believe there is something wrong, as the results do not match

col = size(X,2);
M1 = bsxfun(@minus,X(:), X(:).'); % square form
M2 = theta.*abs(reshape(M1.',[3 col*n*n]).');
M3 = sum((M2(1:10,:)).^p,2)^(1/p);
Mats=squareform(M3)+eye(n);

Any help will be highly appreciated.

Thank you.

Answer 1

[Bruno Luong]

[n,m] = size(X); % do not use length
Xi = reshape(X,[n,1,m]);
Xj = reshape(X,[1,n,m]);
tt = reshape(theta,[1,1,m]);
Mat2 = sum(tt.*abs(Xi-Xj).^p,3).^(1/p);
Mat2(1:n+1:end) = 1;

[Bruno Luong]

If your MATLAB doesn't support auto expansion you need to replace

Mat2 = sum(tt.*abs(Xi-Xj).^p,3).^(1/p)

by two ugly bsxfun statements.

[Gabriele Dessena]

Thank you very much. It works very well.