Ignoring NaN in a mean only if there is a singal NaN

I would like to determine the mean of a set of cells in a matrix but I would like to have the mean be an NaN if there is more than one NaN value in the original data set.

for example if x = [2 NaN 4] then the mean would be 3 if x = [2 NaN NaN] then the mean would be NaN

so far I have been using

nanmean(x(:,1:3),2)

Is there a logical string I could use that would allow me to only ignore the NaN if there was a single NaN in the row?

Answer 1

[Walter Roberson]

One liner:

nanmean(x,2) + 0 ./ (sum(isnan(x),2) < 2)

I suggest you work through it to figure out how it works: it is not obvious (except perhaps to old-timers.)

[Oleg Komarov]

I always forget the trick 0/0.

Answer 2

[Geoff]

Compute the mean with nanmean as usual, and then check for multiple NaNs:

You can count the NaN values in an array like so:

sum(isnan(x))

But you want to do this on a row basis. How about this:

means = nanmean(x, 2);
bad = arrayfun( @(row) sum(isnan(x(row,:))) > 1, 1:size(x,1) );
means(bad) = NaN;

[Oleg Komarov]

sum(isnan(x),2) to sum along columns

[Geoff]

Oh, that makes that big convoluted call a bit pointless then! =) Cheers. I am probably a bit heavy-handed on arrayfun sometimes!