How do you take the nanmean of a 3D matrix?

I have ten matrices that I want to average, but some have NaNs in them. They are of the dimension 64x128x12. How do I get the mean of the ten without including NaN in the calculation and still maintaining the dimensions of 64x128x12?

I have tried nanmean and nanmean2 to no luck.

Thank you!

[Adam]

mean in which dimension? Or just a single scalar mean for the whole lot?

[Walter Roberson]

Question says "mean of the ten"

[HollyA]

A full mean of the ten, similar to if I just did a mean mannually (add them up and divide by 10) but where some have NaN, to ignore that and continue to calculate the mean of the others as a number?

I hope this makes sense?

[Adam]

Well, 'mean of the ten' could still mean take the mean value of all data in all 10 matrices, but from your clarification I assume you mean the end result should be 64x128x12 too.

Walters answer of stacking them into a 4-d 64x128x12x10 matrix and taking the nanmean (or simply 'mean' with an 'omitnan' flag which is the same thing anyway) seems fine to me...

Answer 1

[Walter Roberson]

means_of_10 = nanmean( cat(4, M1, M2, M3, M4, M5, M6, M7, M8, M9, M10), 4);

[HollyA]

I tried this but it seemed to give me a very odd value and not quite work. Thank you for your answer though!

[Walter Roberson]

To confirm: if there was (say) 2 nan out of 10, you would want the others totaled and divide by (10-2) = 8? Or would you want to divide by 10?

You can code more explicitly as:

temp = cat(4, M1, M2, M3, M4, M5, M6, M7, M8, M9, M10);
nans = isnan(temp);
temp(nans) = 0;
means_of_10 = sum(temp, 4) ./ (size(temp,4) - sum(nans,4));

If this does not give you the results you expect then you should check your values more closely.