To find the average when there is NaN

I want to find the average of each pairs. ie. avg of 2 &4, avg of NaN and 5. Answer should be 3 and 5. For the second pair, code should eliminate NaN, so avg should be 5. That's where I have the problem. Following is the code I wrote. Thanks.

    A=[2 4 NaN 5];

    j=1;
    for i=1:2:4
        B=A(i:i+1);
        C=~isnan(B);
        test(j)=mean(A(C));
        i=i+1;
        j=j+1;
    end

Answer 1

[Chad Greene]

Hi Thishan,

On newer versions of Matlab, use

 test(j)=mean(A(C),'omitnan');

If you have an older version of Matlab and you have the Statistics Toolbox, use

 test(j)=nanmean(A(C));

Alternatively, if neither of those options work for you,

tmp = A(C); 
test(j)= = mean(tmp(isfinite(tmp)));

[Thishan Dharshana Karandana Gamalathge]

All of them give the same error i had. test = 3.00 4.00

[Walter Roberson]

mean(reshape(A,2,[]), 'omitnan')

[Thishan Dharshana Karandana Gamalathge]

Thanks. It works.

[Image Analyst]

It will work for A with an even number of elements, not for something like this A=[2 4 NaN 5 6 8 9]; However if you need it to work with an odd number of elements, you can use the code in my Answer, which will work regardless of the number of elements.

[Walter Roberson]

The user asked about "pairs", which does not make sense if there are an odd number of elements.

[Image Analyst]

Sure it makes sense. You can move along the vector an element at a time taking the means of pairs, no matter whether there are an odd or even number of elements. Just look:

m = [1,2,3,4,5]
pairMeans = conv(m, [.5,.5], 'valid')
m =
     1     2     3     4     5
pairMeans =
          1.5          2.5          3.5          4.5

(Note the above does not handle nans in the desired way).

[Walter Roberson]

However, the user's for loop increments by 2, so that is not the pairs they meant. They also indicated they expected exactly 2 results, not 3.

Answer 2

[Image Analyst]

This will work for any length of A, even odd numbers:

A=[2 4 NaN 5 6 8 9]; % 7 elements.
kernel = [1,1];
sumA = conv(A, kernel, 'same')
nonNaNA = ~isnan(A)
denom = conv(nonNaNA, kernel, 'same')
output = sumA ./ denom
output(isnan(output)) = [] % Remove nans.