Choosing the role of NaN elements in the sum environment of matrices

Hello everyone

I would like to know what would be the best way to add element by element two matrices, a and b, which, for simplicity, could be defined as:

aa=[1 2; 3 NaN];
bb=[NaN 1; 2 NaN];

adding some conditions: i) if two elements in position (i,j) are different from NaN, add them and divide by the number of elements in that position other than NaN (for example, for position (2,2): (2+1)/2, and for position (1,1): 1/1), and ii) if two elements in position (i,j) are equal to NaN, the sum goes to NaN. So I was wondering which will be the an efficient way to do aa+bb under the i) and j) conditions to obtain:

cc=[1 3/2; 5/2 NaN];

Answer 1

[Chunru]

aa=[1 2; 3 NaN];
bb=[NaN 1; 2 NaN];
cc = mean(cat(3, aa, bb), 3)

cc = 2x2

NaN 1.5000 2.5000 NaN

[Stephen23]

aa = [1,2; 3,NaN];
bb = [NaN,1; 2,NaN];
cc = mean(cat(3, aa, bb), 3, 'omitnan')

cc = 2x2

1.0000 1.5000 2.5000 NaN

[Chunru]

Yes. It should be with 'omitnan' option.