how to calcuate mean with NaN

Dear Friend,

There are two matrixs with the same size, there are some NaNs in each matrix. I want to calculate the mean of corresponding datapoints, which might contain NaN. for example, A(1,2)=NaN, B(1,2)=3, I need the average of the sum of 3+NaN divided by the effective number of data points, here is 1 since A(1,2) is NaN, to be 3. If A(2,2)=2,B(2,2)=5, I need the average to be (2+5)divided by 2, which equals 3.5 since neither A(2,2) and B(2,2) contains NaN. Is there a way to achieve this goal without using a for loop? thanks

Answer 1

[Sven]

If you have the Statistics Toolbox, just replace calls to mean() with calls to nanmean().

If you don't have it, you can do the same by downloading the File Exchange contribution nansuite

Here's an example:

A = reshape(1:4,2,2);
B = A + 10;
A([1 3]) = nan % Set some NaNs
B([1 2]) = nan
yourMean = nanmean(cat(3,A,B),3)

Which results in:

A =

     NaN   NaN
       2     4

B =

     NaN    13
     NaN    14

yourMean =

     NaN    13
       2     9

[Peihong]

Hi, Sven, thanks a lot. It's exactly what I need.

Answer 2

[Jan]

Emulating NANMEAN without the statistics toolbox is easy:

index    = isnan(X);
X(index) = 0;
M        = sum(X, 2) ./ sum(index, 2);

[Esen S.]

I belive there is a tilde missing in the last line.

It should be M = sum(X, 2) ./ sum(~index, 2);

Answer 3

[Shashank Prasanna]

You can use the NANMEAN function part of the Statistics Tbx:

http://www.mathworks.com/help/finance/nanmean.html

http://www.mathworks.com/help/finance/financial-times-series-descriptive-statistics.html