How can I apply arrayfun with 2D matrix inputs?

I want to accelerate the following code.

B=zeros(N,M,M);
for n=1:N
  B(n,:,:)=inv(A(n,:,:));
end

where the size of A is N x M x M. Arrayfun doesn't support such a matrix input. So I would like to know the way to accelerate it.

Thank you in advance!

[Andrei Bobrov]

It is unlikely that the arrayfun will accelerate.

[Daichi]

Thank you. I see. Is there any way to handle this kind of processing?

Answer 1

[Andrei Bobrov]

B = permute(A,[2,3,1]);
eye3 = eye(3);
for jj = 1:size(B,3)
    B(:,:,jj) = B(:,:,jj)\eye3;
end
B = permute(B,[3,1,2]);

or within cellfun:

B = permute(cell2mat(cellfun(@(x)x\eye(3),...
                                  num2cell(permute(A,[2,3,1]),[1,2]),'un',0)),[3,1,2]);

[Daichi]

I see that cellfun is available in this case. But I eventually found that this kind of processing cannot be accelerated even if we use cellfun as you described. I guess num2cell and cell2mat may require some computational costs. Fortunately, the first one using backslash operator was faster. Anyway, thank you so much for your kind answer.

[Andrei Bobrov]

Within arrayfun:

B = permute(A,[2,3,1]);
Bc = arrayfun(@(ii)B(:,:,ii)\eye(3),reshape(1:size(B,3),1,1,[]),'un',0);
B = permute(cell2mat(Bc),[3,1,2]);