Remove the loop from a sum of index expression

Consider the following simple example:

% Example setup
n = 5;
m = 10;
M = randn( m, m );
J = randi( n, m, m );

% Code to simplify
V = zeros( n, 1 );
for i = 1 : n
    V( i ) = sum( M( J == i ) );
end

Is it possible to replace the loop by a one line expression?

For the sake of answering this, assume that I do not particularly care about either efficiency, or readability. I really just want a one line solution.

The actual code does not have random M and J, so I am also not interested in solutions based on these particular distributions.

[Jiri Hajek]

Hi, your example code produces a vector in each step of the loop, is that a typo? Please clarify, what is the desired result.

[Tom Holden]

No it doesn't. The sum collapses the vector.

Answer 1

[Stephen23]

Hiding the loop even more (really everything uses loops, even vectorized code):

% Example setup
n = 5;
m = 10;
M = randn( m, m );
J = randi( n, m, m );
% Code to simplify
V = zeros( n, 1 );
for k = 1:n
    V(k) = sum(M(J==k));
end
V

V = 5×1

-8.9895 -0.9206 -1.5965 0.6393 -3.7407

% Simpler code:
V = accumarray(J(:),M(:),[n,1])

V = 5×1

-8.9895 -0.9206 -1.5965 0.6393 -3.7407

[Tom Holden]

Somehow I have never encountered "accumarray" before. But it is perfect here.

Answer 2

[Torsten]

Here is a loop in disguise:

n = 5;
m = 10;
M = randn( m, m );
J = randi( n, m, m );
V = arrayfun(@(i) sum( M( J == i ) ),1:n)   

[Tom Holden]

A fair answer! It still feels like there ought to be a solution that is less explicitly a loop than this.