How to set an array identifier within code

Hi,

I am trying to write a snippet that takes an input array of any size (I will refer to it as A with N dimensions) and a column or row vector v, that returns an array B having N+1 dimensions, which elements along the (N+1)th dimension are the products A*v(i) (where v(i) are the individual scalar elements in v).

If A is a 2D matrix, it can be done this way:

A=rand(4,5);
v=rand(1,3);

B=zeros([size(A) length(v)]);
for k=1:length(v)
    B(:,:,k)=v(k)*A;
end

This works fine but since I need to write as many ' : ' as there are dimensions in A, this is not very versatile.

So I am looking for a way to change this argument as the code executes to adapt to the number of dimensions in A. I have tried the following:

A=rand(4,5);
v=rand(1,3);

bit=sprintf(':,'); %Define the string element to be repeated N times in the argument
arg=bit; %Initiate the argument string
for c=1:(length(size(A))-1) 
    arg=strcat(arg,bit) %Concatenate as many times as necessary
end
arg=strcat(arg,'k') %Append with the letter k
%In this example, arg is ':,:,k'

B=zeros([size(A) length(v)]);
for k=1:length(v)
    B(arg)=v(k)*A;
end

But it doesn't work...

Answer 1

[Guillaume]

Trying to build the indices as text is never going to be efficient. What you want can be trivially obtained:

%A: an N-d matrix
%v: a vector
%B: an (N+1)-d vector where B(..., k) = B(...) .* v(k). Dimensions 1:N have same size as A, dimension N+1 is numel(v)
B = A .* reshape(v, [ones(1, ndims(A)), numel(v)]);  %move v in the N+1 dimension and multiply

[Julien Lopez]

I think I see where you're coming from but I get a "Matrix dimensions must agree" error with the elementwise multiplication...

[madhan ravi]

B = bsxfun(@times,A,reshape(v, [ones(1, ndims(A)), numel(v)]));

[Guillaume]

Up there, in the right corner of the webpage there's a field for you to enter the version of matlab that you're using. It's important to fill out if you're on an old version.

My guess is that you're using a version prior to R2016b which does not have implicit expansion. In that case,

B = bsxfun(@times, A, reshape(v, [ones(1, ndims(A)), numel(v)])));

The code in my answer is guaranteed to never generate a Matrix dimensions must agree error in R2016b and later.

[Julien Lopez]

Oh alright, sorry about that, the computer I'm using is indeed running on R2016a. Thanks a million to both of you!