Arrayfun with multidimensional input

I have a function that take three inputs. The first two inputs (say 'a' and 'b') are integers, and the third input (M) is a 3D matrix. The function performs some computations on M using the integer inputs a and b (eg. dataOut = a + b .* (M)). Now let's say we have several values for a and b, stored as vectors while M is constant. I want to vectorize my function using arrayfun, but arrayfun requires the inputs (a,b,M) to be of the same size. I tried passing M as a structure that matches the size of a and b, but that doesn't work either. Eventually, if arrayfun works for this case, I would store a, b and M as gpuArrays and would like to run the code on GPU. Any help is appreciated.

[James Tursa]

If "a" and "b" are vectors, what would the desired output be? A 4D array?

[Kishore Rajendran]

Dear James Tursa. Thanks for your reply. Yes, the desired output will be a 4D array. Suppose a and b are (1 x 100) vectors, and M is 200 x 200 x 200, then the desired output will be 200 x 200 x 200 x 100. I can also do away with cell array as output which will be a (1 x 100) cell, and each cell element will have a 200 x 200 x 200 matrix.

Answer 1

[Edric Ellis]

You can do this by "binding" in M to your function handle, using an anonymous function, like so:

a = 1:10;
b = rand(1, 10);
M = rand(4);

arrayfun(@(aa, bb) norm(aa + bb.*M), a, b)

This sort of approach can work for gpuArray too, but beware that the functions that you are allowed to call inside gpuArray arrayfun is restricted to this list.

[Kishore Rajendran]

Thanks Edric. Much appreciated. My function is a little more complicated than the norm you had used in your example. I will give it a try using my custom function and see if that would work.