Canonical Correlation Analysis - canoncorr function in matlab

I am just starting with CCA, I am trying to explore it using the matlab function, but I guess I am stuck understand the result I get using canoncorr function.

[A,B,r,U,V] = canoncorr(X,Y)

My understanding is that I would get U=A'*X, and Y=B'Y, but doing that with the return type does not give me the expected result.

I tried looking at the code for canoncorr. It has something like

    if nargout > 3
       U = X * A;
       V = Y * B;
    end

I am not sure, why U is being calculated like that, as opposed to definition here http://en.wikipedia.org/wiki/Canonical_correlation

Answer 1

[Tom Lane]

Try comparing V with the product

bsxfun(@minus,Y,mean(Y))*B

The first factor is just Y with the column means subtracted. Alternatively, you can compute Y*B and subtract the column means of the result.