Azzera filtri
Azzera filtri

Multiplying certain dimensions of two 3D matrices

5 visualizzazioni (ultimi 30 giorni)
Hi
I've got two matrices, one 30x30x50 and one 30x30x80
I would like to multiply them in such a way that the result is a 50x80 matrix ie. an implicit sum over all of the 30x30 common elements.
eg. A*B = C
C(k,l) = A(i,j,k)*B(i,j,l) summed over all i, j
Is there a simple way to write this? I would rather avoid complications using "reshape".
Thanks, Mike

Risposta accettata

Andrei Bobrov
Andrei Bobrov il 14 Mag 2012
Idea of Sean de Wolski
C = squeeze(sum(sum(bsxfun(@times,a,permute(b,[1 2 4 3])))))
or
s = size(a)
C = squeeze(sum(bsxfun(@times,reshape(a,s(1)*s(2),[])',reshape(b,1,s(1)*s(2),[])),2))
  1 Commento
Andrei Bobrov
Andrei Bobrov il 14 Mag 2012
Hi Michael! Thank you for "accpted", but Teja's variant is better than my.

Accedi per commentare.

Più risposte (2)

Teja Muppirala
Teja Muppirala il 14 Mag 2012
A = rand(30,30,50);
B = rand(30,30,80);
C = reshape(A,900,[])'*reshape(B,900,[]);
  4 Commenti
Michael
Michael il 14 Mag 2012
I'll use this one, thanks everyone
Michael
Michael il 14 Mag 2012
I have one more related question,
I'm also trying to compute D = C*A, I think A must be transposed for this to make sense: the (i,j,l) element of D is = (k,l) of C times (i,j,k) of A summed over k. I've experimented a bit and found this,
reshape(reshape(A,900,[])*C,30,30,[]);
Is it the best method? I haven't found a general rule to do these automatically yet and I'm not sure if the 'outer' reshape is preserving the form of my 30x30 matrix and not inadvertently flipping it about a diagonal.
Thanks very much for any help

Accedi per commentare.


Comrun Yousefzadeh
Comrun Yousefzadeh il 17 Giu 2020
Modificato: Comrun Yousefzadeh il 17 Giu 2020
I have a related questoin. I'm trying to multiply a two dimentioanl array to a specific dimension of a 3-D matrix. For example, A=2000*2000*72, B=72*3 and I'm trying to reach C=2000*2000*3. In other words B should be multiplied to the 3rd dimension of A and repeated (element wise) for 2000*2000 points in the first two dimensions. Currently I'm using a loop like this:
for ii=1:size(A,1)
for jj=1:size(A,2)
A1=A(ii,jj,:);
A1=A1(:);
C1=B'*A1; % C1 has dimension of 3*1
C2=D*C1; %D is a 3-3 matrix therefore, C2 has the same dimension as C1
C(ii,jj,:)=C2(:,1);
end
end
What is the better way to avoid the loop?
Thanks,

Categorie

Scopri di più su Resizing and Reshaping Matrices in Help Center e File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by