Fastest way to substitute elements in a matrix at given positions?

Good evening,

suppose I have an array A with size (n,m) and an array B with size (n,l), with l<m.

Suppose elements of array B point to elements of array A in the following fashion:

• B(x,y) --> A(x, B(x,y))

What is the fastest way to substitute all elements of A that are pointed by elements of B with a given value? At present I am using for loops but I guess they are not very efficient.

Also, would it be different if matrix A would be generated by repetition of the same known row n times?

Thanks a lot for any help or hint you are willing to give!!

Answer 1

[Matt J]

idx = sub2ind([n,m], repmat((1:n).',1,l)  ,B);

A(idx)=value;

[Torquato Garulli]

That really helped a lot. Thank you very much!

If you wish, could you explain why is it that this is faster?

Thanks anyway!

[Matt J]

Faster than what?

[Torquato Garulli]

Than using loops

[Matt J]

It depends on what the code for the loop looks like. Generally speaking though, Matlab's JIT compiler can do some things to optimize M-coded for-loops, but is limited by the complexity of operations done inside the loop and your Matlab version, of course.

If you can avoid M-Coded for-loops, as my solution does, it often leads to better performance because it doesn't rely on the JIT. It is just running pre-compiled C/C++ code internally.

[Torquato Garulli]

I see!

Thanks a lot!