Sum matrix with repeating indices

Hi! I have one matrix, A, that has one column of identifier values (with repeats and some missing numbers) and a column of values. I'd like to produce a column of all of the unique values and then another column of their sum. For example, if A is as follows, i want to produce B.

A = [1  2
     1 -3
     2  0
     3  3
     5  6
     2 -4
     4  0.5
     7  9
     7  2]
% I want to produce B
B = [1 -1
     2 -4
     3  3
     4  0.5
     5  6
     7  11]

Answer 1

[Stephen23]

A simple MATLAB approach:

A = [1,2;1,-3;2,0;3,3;5,6;2,-4;4,0.5;7,9;7,2]

A = 9×2

1.0000 2.0000 1.0000 -3.0000 2.0000 0 3.0000 3.0000 5.0000 6.0000 2.0000 -4.0000 4.0000 0.5000 7.0000 9.0000 7.0000 2.0000

[G,U] = findgroups(A(:,1));
M = [U,accumarray(G,A(:,2))]

M = 6×2

1.0000 -1.0000 2.0000 -4.0000 3.0000 3.0000 4.0000 0.5000 5.0000 6.0000 7.0000 11.0000

[Stephen23]

"What is your definition of simple?"

Hmmm... off the top of my head, something like this: simple code uses functions and syntax whose intent is clear, making understanding the code intent easier to understand when it is read through by regular users of that language, and generally making the code more efficient. Simple code utilizes existing routines effectively (i.e. does not reinvent the wheel), and only specializes sufficiently to solve the task at hand.

"Where can I find a reference to this?"

When I search for "simple code" using a major internet search engine it returned 2690000 results. I read a few of them just now, this one I think gives a reasonable summary (it also happens to match very closely with my own definition of "simple" given above):

https://medium.com/javascript-scene/the-secret-of-simple-code-a2cacd8004dd

The concept of abstraction seems to be very commonly associated with simple code:

https://www.wolfram.com/language/elementary-introduction/2nd-ed/46-writing-good-code.html

I.e. understanding the underlying model and knowing how that can be best represented in code.

"How is it measured?"

I can think of several ways to measure it:

• syntax metrics (e.g. cyclomatic complexity, number of operations, etc)
• use of existing routines and paradigms as much as reasonable (related to abstraction)
• use of methods that are generalized as much as reasonable, e.g. type/data agnostic
• metrics related to maintainability, readability, etc.

Note that simpler does not mean shorter.

[Steven Lord]

Instead of using findgroups and accumarray I would probably use groupsummary.

A = [1,2;1,-3;2,0;3,3;5,6;2,-4;4,0.5;7,9;7,2];
[values, groups] = groupsummary(A(:, 2), A(:, 1), @sum);
results = [groups, values]

results = 6×2

1.0000 -1.0000 2.0000 -4.0000 3.0000 3.0000 4.0000 0.5000 5.0000 6.0000 7.0000 11.0000

Answer 2

[Voss]

One perfectly valid MATLAB approach:

A = [1  2
     1 -3
     2  0
     3  3
     5  6
     2 -4
     4  0.5
     7  9
     7  2];

[uA,~,jj] = unique(A(:,1));
n_uA = numel(uA);
B = [uA zeros(n_uA,1)];
for ii = 1:n_uA
    B(ii,2) = sum(A(jj == ii,2));
end

disp(B)

    1.0000   -1.0000
    2.0000   -4.0000
    3.0000    3.0000
    4.0000    0.5000
    5.0000    6.0000
    7.0000   11.0000

% % I want to produce B
% B = [1 -1
%      2 -4
%      3  3
%      4  0.5
%      5  6
%      7  11];