Summing two vectors with NaNs

I need to sum two vectors that might contain NaNs. If one vector has a Nan and the other has a number, I want the sum to be the number.

The sum function doesn't have this option.

A = [ 1 2 3 NaN NaN NaN 7 8 ]';
B = [ 1 1 1 1   NaN NaN 1 1 ]';
C = sum( [A,B], 2, 'omitnan' )
C =

     2
     3
     4
     1
     0
     0
     8
     9
C = sum( [A,B], 2, 'includenan' )
C =

     2
     3
     4
   NaN
   NaN
   NaN
     8
     9

I want the result to be this:

C =

     2
     3
     4
     1
   NaN
   NaN
     8
     9

Any suggestions how to achieve this?

Answer 1

[dpb]

A = [ 1 2 3 NaN NaN NaN 7 8 ]';
B = [ 1 1 1 1   NaN NaN 1 1 ]';

Check which rows are ok...

isOK=isfinite(A)|isfinite(B);
C=sum([A,B],2,'omitnan');
C(~isOK)=nan

C = 8×1

2 3 4 1 NaN NaN 8 9

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

Could encapsulate in a little function to make clean at top level...

mysum=sumeither(A,B)

mysum = 8×1

2 3 4 1 NaN NaN 8 9

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

●

function s=sumeither(A,B)
% returns sum of finite elements of vectors A,B
  assert(isvector(A)&&isvector(B),'Both inputs must be vectors')
  assert(numel(A)==numel(B),'Both inputs must be same length')
  M=[A(:) B(:)];          % catenate to array, be sure are columns
  s=sum(M,2,'omitnan');   % add up the finite elements
  s(all(isnan(M),2))=nan; % mark the missing rows
end

Answer 2

[Thorsten]

s = sum([A,B], 2, 'omitnan');
s(isnan(A) & isnan(B)) = nan;