Column multiplication of two 2D matrices.

Hi there,

I have two quite large matrices with the following sizes:

A = rand(94037,2240) B=rand(2240,2240)

I would like to mutiply each column of A by the transpose of each column of B (bascially two vector multiplication). Therefore each mutiplcation gives me a 2D matrix wih size (94037,2240) and I would like to store the result of each multiplication in a 3d MATRIX.

The thing is I can do this with for-loop as show below but its very slow and in the middle of operation matlab exist, because its quite heavy.

S = zeros(94037,2240,2240);

for i=1:1:2240

S(:,:,i) =A(:,i)*(B(:,i))';

end

Is there anyway I can do this without using for loop to speed up the process ?

Thanks for the help.

Best,

Nikan

[Matt J]

Outer products are often both an inefficient and an avoidable step in many computations. What do you plan to do with S once you've computed it?

[Nikan Fakhari]

I am planning to take sum along the third dimension

Answer 1

[Matt J]

I am planning to take sum along the third dimension

That would simply give you A*B.'

Answer 2

[Matt J]

I doubt you will be able to do the computation with the matrix sizes you've indicated. The result will consume 3.5 TB of RAM in double floating point. However, if you can scale down the matrix sizes, then a loop-free solution is as follows,

[ma,na]=size(A);
[mb,nb]=size(B);

S=reshape(A,ma,1,na).*reshape(B,1,mb,nb);

[Bruno Luong]

I guess there is a typo. The right command has no brakets

S = reshape(A,ma,1,na) .* reshape(B,1,mb,nb);

[Matt J]

Right!