Remove NAN from cell array, and calculate the average of each column through each cell

I have a 1 x 17 cell array called M, with each cell containing a 16 x 27 Matrix. I need to get rid of NaN in each cell. I tried this code:

for i=1:length(M) ind = isnan(M{1}); M{i}(ind)=[]; end

But the cells are reshaped to rows (1x240) and I need the expected result to be a p*q Matrix. Once the expected result is obtained, I have to go through each cell and calculate the average of each column q Any help please?

[Punit]

First of all

ind = isnan(M{1});

should be ind = isnan(M{i});

Also, try replacing M{i}(ind) = 0;

then sum(M(:,q))/length(find(~ind(:,q))) would be the average of column q

[Guillaume]

What does 'get rid of NaN' mean? Obviously, you need something in that matrix entry where the NaN is: a number, NaN, Inf or +Inf are the only possibilities.

If you remove the matrix element, then the matrix cannot be square anymore.

Answer 1

[Guillaume]

As of version 2015a, mean can ignore NaNs (in earlier version you had to use nanmean in the stat toolbox).

So the simplest way to calculate the mean is with:

colmeans = cellfun(@(m) mean(m, 'omitnan'), M, 'UniformOutput', false);
celldisp(colmeans);

Even simpler, would be to just store all your matrices as a 3D matrix (of size 16 x 27 x 17) and avoid cell arrays entirely:

M = cat(3, M{:});
colmeans = mean(M, 'omitnan')

[F Z]

i got the 2014 version so that 'omitnan' isn't available. i tried the following code to remove NAN rows and columns

for k=1:length(M)
    R=M{k};
    R=R(4:9,2:22); %%to remove all NANs 
    T{k}=R;
end

T = cat(3, T{:});
colmeans = mean(T);

it seems to work. Thank you a lot for your help

[Guillaume]

A simpler version of your code would be:

T = cat(3, M{:});
colmeans = mean(T(4:9, 2:22));

[F Z]

brilliant!! Thx a lot