Error using sum function

I have a 26x26 matrix 'A' of type double. I want to sum all of its elements using S = sum(A, 'all').

I keep getting the following error:

Error using sum

Invalid option. Option must be 'double', 'native', 'default', 'omitnan' or 'includenan'.

Can someone help me with this? My matrix is of the correct type, so I am not sure what's wrong.

[Awais Ali]

use S=sum(sum(matrix))

[Steven Lord]

Please show us the output of the version function. I agree with the other posters that you probably are using a release prior to the introduction of that option.

[Seangleng Khe]

I had this problem too. I cannot use this command. Yet I use sum(sum(c)). It will give you the same result.

[Walter Roberson]

@Seangleng Khe which MATLAB release are you using, and what is class(c) ?

[Seangleng Khe]

@Walter Roberson What I mean is that I face this problem too. So instead of sum(A,'all'), which does not work for me, I use sum(sum(A)). There is nothing relating to class(c).

[Stephen23]

"It will give you the same result."

No, not in general:

format short G
A = rand(4,3,2);
sum(sum(A))

ans = 
ans(:,:,1) =

       4.1374


ans(:,:,2) =

       4.1502

sum(A,'all')

ans = 
       8.2876

sum(A(:))

ans = 
       8.2876

[Walter Roberson]

@Seangleng Khe

Please show us the output of the following command:

version

[Paul]

sum(sum(X)) is not guaranteed to give the exact same result as sum(X,'all'), even for two-dimensional X

rng(101)
X = rand(100);
isequal(sum(X,'all'),sum(sum(X)))

ans = logical
   0

These are equal for this case, but is it guaranteed for all X?

isequal(sum(X,'all'),sum(X(:)))

ans = logical
   1

[DGM]

To extend this to what Walter said about class:

% i'm avoiding rand() hopefully avoid any version/platform
% sensitivity in generating the test array, mostly because i wanted 
% to run the test in versions prior to the introduction of rng()
X = magic(1000)/3; % X is double

SA = sum(X,'all');
abs(SA - sum(sum(X))) % error scales with data

ans = 3.0518e-05

abs(SA - sum(X(:))) % equal

ans = 0

X = single(X); % X is single
SA = sum(X,'all');
abs(SA - sum(sum(X))) % error is not insignificant

ans = single 278528

abs(SA - sum(X(:))) % still equal

ans = single 0

So for a 2D input, sum(sum(X)) does not give the same result, and the difference between the methods does depend on the array class, and it also depends on the version (and maybe platform) as well. I get different results on every version I've tested. But in all cases, sum(X,'all') and sum(X(:)) yield identical results.

We can question whether it's prudent to blindly try taking the sum of large single-precision arrays, but that's not really the point.

[Bruno Luong]

The sum is never guaranteed to be indentical even for the same input, same (apparent) command, same PC, same version

X=rand(1,1e6);

n=maxNumCompThreads(1); s1=sum(X,'all');
maxNumCompThreads(n); sn=sum(X,'all');

isequal(s1,sn)

ans = logical
   0

s1-sn

ans = -2.3283e-10

[DGM]

Ah.

[Paul]

I assume sum(X,'all') is implemented as sum(X(:)) at the intermediate interface level before the sum hoerachy and low-level CPU instruction occurs, so I think yes, they should return the same result.

Answer 1

[madhan ravi]

sum(A(:)) % if it doesn't work either upload your matrix as .mat file

[Nour Aburaed]

This is one way around it, but I still would like to know the reason for that error

[madhan ravi]

Edited:

Maybe because your using an older version of Matlab , I believe 'all' (vector dimension argument) was released in 2018b. See https://www.mathworks.com/help/matlab/release-notes.html?rntext=%22vector+dimension%22&startrelease=R2015aSP1&endrelease=R2018b&groupby=release&sortby=descending&searchHighlight=%22vector+dimension%22 . What release are you using ? upload your matrix as .mat file to experiment (to see if I can reproduce that error) .

[Nour Aburaed]

I am using the latest release

[Stephen23]

"I believe 'all' name value pair... "

It is not a "name-value" pair, it is simply an optional input argument. For example, see textscan for examples of optional inputs which must be given as name-value pairs:

sum(...,'all') % not a name-value pair.
textscan(...,'Delimiter',';')
%             ^^^^^^^^^       name
%                        ^^^  value

[John D'Errico]

Actually, you cannot be using the newest release. Currently, the newest release is R2018b.

This capability:

S = sum(X,'all') sums all elements of X.

was introduced in R2018b. It did not exist in R2018a. (I have both versions active on my computer. So I verified that fact.)

You may think you are using that release, but if sum fails, producing the error described, then you are using an older release.

For example, in R2018b, you can do this:

A = rand(26,26);
sum(A,'all')
ans =
          336.422314041605

That will fail in any older release.

You can check what release you are running by simply using ver. At the command line in MATLAB, type

ver

In R2018b, you should see something like:

-----------------------------------------------------------------------------------------------------
MATLAB Version: 9.5.0.944444 (R2018b)
MATLAB License Number: ********
...

[madhan ravi]

Thank you very much John D'Errico appreciate it.

@Nour respond to John D'Errico's comment and let us know..

[Image Analyst]

It's possible he's using the latest student release. I heard that in most years there is no "b" student release, just an "a" release. But that wouldn't have 'all' in it, however (:) will always work.

[madhan ravi]

But sir Image Analyst I’m using student version too , it works in my case though.

[Steven Lord]

Image Analyst: clicking on the "Buy MATLAB and Simulink Student Suite" button on this page brings up the store page for the Student Version of release R2018b.