Force nansum to equal NaN (and not 0)

Hi Mathworks

I have had a lot of luck with previous questions, so I will try once more. The question have sort of already been answered before here, but I can seem to get the suggestion to work:

https://se.mathworks.com/matlabcentral/answers/262694-how-to-force-nansum-nan-nan-nan-not-0

Basically, I have a matrix of some numbers (and a lot of NaNs) that I want to multiply by a similar matrix and take the nansum (line 168-171) as:

for i                  = 1:(size(retSize1Value1,1))
     tmp_i             = nansum(omega(i,:).*retSize1Value1(i,:));         
     returnMarket(i,1) = tmp_i(1); 
end

returnMarket(all(isnan(omega)&isnan(retSize1Value1),1)) = NaN;

I have tried the suggestion in the other post (line 173) after the for loop.

Later I will take the mean, so the issue is, that if nansum is a product of NaNs Matlab fill out the value to zero and not NaN!

I hope you will save me once more.

All the best,

Christoffer

Answer 1

[Spencer Chen]

Try:

returnMarket(all(isnan(omega)&isnan(retSize1Value1),2)) = NaN;

You need to run all() on dimension 2 instead of dimension 1.

A few more tips on your code:

1. You will find your problem easier to debug if you reduce the complexity of your expressions. i.e. Code the above line in 2 steps:

nanmask = all(isnan(omega)&isnan(retSize1Value1),2);
returnMarket(nanmask) = NaN;

2. Your for loop can be easily converted into matrix operations, which tends to improve speed and helps you to think more Matlab-like. I believe the below produces an equivalent result to your for loop:

tmp_i = omega .* retSize1Value1;
returnMarket = nansum(tmp_i,2);

Blessings,

Spencer

[Christoffer Benneballe]

Thank you, Spencer!

It works now - really nice!

And appreciate your suggestions - I cannot seem to get the second part to function as suggested under 2.

Tried to implement it as follows and also tried to comment out the last term. I just get returnMarket = 0.

for i                  = 1:(size(retSize1Value1,1))
     tmp_i             = nansum(omega(i,:).*retSize1Value1(i,:));
     returnMarket      = nansum(tmp_i,2);
     returnMarket(i,1) = tmp_i(1); 
end