Sum of columns in a mxn Matrix on matlab

function sum =column_sum(A,cl)

    sum=[];
    [m,n]=size(A);
    if cl==n
        sum=A(cl,:)';
    else
        s=column_sum(A,cl+1);
        for i=1:n
            sum=[sum;s+A(cl,i);];
        end
    end
end

This is my code. cl stands for current line. But im getting the error:

error: 'A' undefined near line 3, column 12
error: called from
column_sum at line 3 column 6
>>

[Matt J]

I'm getting no such error.

A=rand(4);

column_sum(A,3)

ans = 16×1

0.3737 0.2950 0.9750 0.8340 0.4587 0.3800 1.0600 0.9190 1.1090 1.0303

function sum =column_sum(A,cl)

    sum=[];
    [m,n]=size(A);
    if cl==n
        sum=A(cl,:)';
    else
        s=column_sum(A,cl+1);
        for i=1:n
            sum=[sum;s+A(cl,i);];
        end
        
    end

end

[test test]

Maybe the error has something to do with file paths then. Does it actually do what i want? Sum of the columns of the mxn matrix?

Answer 1

[Image Analyst]

Why have a function to sum a column? Why not simply do

columnSum = sum(A(:, cl), 1)

[test test]

My teacher gave us this code which is for sum row:

function y=row_sum(A)
[m,n]=size(A)
for i=1:m
    s=0
    for j=1:n
        s=s+A(i,j)
    end
    rs(i)=s
end
y=rs
end

and she wants us to do it for sum columns now.

[Image Analyst]

Just iterate over i=1:m instead of over j.

[test test]

function y=column_sum(A)

[m,n]=size(A)

for i=1:m

s=0

for i=1:m

s=s+A(i,j)

end

cs(i)=s

end

y=cs

end

So like that?

[Image Analyst]

Like this

function [colSums, rowSums] = column_sum(A)
[rows, columns] = size(A);
% Get column sums by summing each column over all rows
colSums = zeros(1, columns);
for row = 1 : rows
    thisRow = A(row, :);
    colSums = colSums + thisRow;
end
% Get row sums by summing each row over all columns
rowSums = zeros(rows, 1);
for col = 1 : columns
    thisCol = A(:, col);
    rowSums = rowSums + thisCol;
end
end

[test test]

Do you know how to edit this in order to find the sum of the entire matrix?

[dpb]

Just take out the reinitialization of the accumulator s inside the loop and don't save the row sum from the instructor's original version. That continues to use the doubly-nested loop.

Of course, outside the pedgogical exercise teaching you to write loops and indexing, "the MATLAB way" when you get to really using MATLAB "in anger" is

S=sum(A,'all');

no loops are needed at all.

You'll see in older code idioms like

S=sum(sum(A));

or

S=sum(A(:));

but those are now anachronisms and should be avoided going forward.

One bad practice in the instructor's code example is that the s vector used to accumulate the row sums is not preallocated -- this won't show up with small arrays, but is poor practice and will become a performance issue when sizes become large and also should preallocate whenever possible. Here that would have been at the beginning of the function before the loop

s=zeros(m,1);       % preallocate a column vector same size as input array # rows

[Image Analyst]

To sum an entire matrix you can use

S = sum(A(:))

A Mathworker told me there is no slowness introduced by using (:).

If you really want the for loop way, you can do it with a single for loop

S = 0
for k = 1 : numel(A)
    S = S + A(k);
end

Or you can use the 'all' option of sum

S = sum(A, 'all');

[dpb]

A Mathworker told me there is no slowness introduced by using (:)."

Agreed, the use of the (:) is not terrible choice as compared to using the 'all' keyword; I could have made that proscription a little less hard. The nested sum(sum()) is an ugly hack still, however, seen far too often even in new code the idiom undoubtedly having been picked up by seeing it in older code or from the self-taught instructor who learned it the same way and never improved their own coding style and continues to pass on the same bad habits (as well as the penchant for "clear all" everywhere, even inside functions).

Sorry about the typo Indeed...I swear the Answers interface doesn't always catch keystrokes when typing...I discover a tremendous number of typos that I would swear I didn't make... :)

Answer 2

[dpb]

    s=column_sum(A,cl+1);

Will crash if pass cl > size(A,1) as there's no error checking for cl being out of range.

As for the Q? "Does it actually do what i want? Sum of the columns of the mxn matrix?" that depends on the actual definition intended. If it is to return simply the sum of a given row in the array A, then all that is needed is

function ans=column_sum(A,cl)       % don't overload the builtin sum method
  m=size(A,1);                      % number rows in A
  assert(m>=cl,'Error:  passed row greater than array size')
  ans=sum(A(cl,:));
end

It's not clear at all beyond that what your expectations are -- by the recursive call you have, as @Matt J's example shows, you're returning a vector the size of the input array.

To return the sum of all rows in the array, just use the optional dim argument in sum as

col_sum=sum(A,2);       % all rows summed by column dimension

If the idea is to only return those from the specified row to the end and not the full array size, then

col_sum=sum(A(cl:end,:),2); % all rows summed by column dimension from row cl:end

or simply

col_sum=sum(A,2);           % all rows summed by column dimension
col_sum=col_sum(cl:end);    % return only from row cl:end

You'll have to 'splain just what it is that you're really after explicitly; it's ambiguous as given.