How to force nansum(NaN+ NaN) = NaN, not 0 ?

Hello,

I need to sum some multidimentional arrays after SPMD processing. This arrays contains also NaNs, but I need to computer a proper mean value of them later. Since:

>> nansum(NaN + NaN)
ans =
     0

This case will affect the result. How to perform such a summing, but sum of ONLY NaNs has to be equal to NaN (not zero!)?

Best regards, Alex

Answer 1

[Stephen23]

Just keep track of where all elements are NaN, and replace their sum with NaN after using nansum. You should be able to adapt this example to help you, it sums along each column:

>> A = [1,2,NaN,NaN;3,NaN,4,NaN]
A =
     1     2   NaN   NaN
     3   NaN     4   NaN
>> B = [6,7,8,NaN;9,NaN,NaN,NaN]
B =
     6     7     8   NaN
     9   NaN   NaN   NaN
>> C = nansum(A+B,1)
C =
    19     9     0     0
>> C(all(isnan(A)&isnan(B),1)) = NaN
C =
    19     9     0   NaN