Replacing elements of multilevel cell arrays

Hello, I have an multilevel cell arrays of meancalc. For example: To reach any of them, I have to write meancalc{1,1}{:,:}, meancalc{1,2}{:,:},meancalc{1,3}{:,:}. I want to reach the Inf values and change them with epsilon. But I couldn't reach them out with meancalc(meancalc{:,:}==Inf) = eps; Thanks in advance.

[Guillaume]

Why eps(1)? If it's just an arbitrary small value, eps(1) is not that small (compared to realmin)

[Sukru Yavuz]

I am calculating the standard deviation of columns. And the values of some columns are identical. That's why the results are 0. After that step, I am using those zeros in a multiplication. That's why I am getting the Inf results. My advisor told me to replace them with epsilon to avoid Inf or NaN or 0s.

[Guillaume]

Well, you of course do whatever you want, but replacing an easily spotted erroneous value (maybe?) by an arbitrary constant value does not sound logical to me. Why eps(1) instead of exp(1) or sqrt(2) or eps(1e-10) or any other random arbitrary value that has no more significance than eps(1) to your algorithm?

Answer 1

[Guillaume]

Assuming that all meancalc{t,u}{v,w} contain matrices, you'll have to use a double for loop:

for outer = 1:numel(meancalc)
   for inner = 1:numel(meancalc{outer})
       innermat = meancalc{outer}{inner};
       innermat(isinf(innermat)) = eps(1);
       meancalc{outer}{inner} = innermat;
   end
end

Note: acellarray{:} will return a comma separated list, so your meancalc{:,:}==Inf would never work.

[Sukru Yavuz]

It works perfectly. Thank you!

Answer 2

[Birdman]

Usage of cellfun might help:

 idx=cellfun(@(x) isinf(x),meancalc)

This will output logical array. Then,

 meancalc(idx)={234};%Inf's are replaced with a random value, change this with eps

[Sukru Yavuz]

What does x stand for? If they stand for meancalc, then I am getting this error: Undefined function 'isinf' for input arguments of type 'cell'.

[Birdman]

Hmm, which version of MATLAB are you using?

[Guillaume]

No cellfun won't help because:

a) you want to do an assignment which is not supported by anonymous functions. You'd need a separate function to do the assignment, at which point you're better off using an explicit loop (in my opinion)

b) the content of the cell array is itself a cell array, so you'd need at least a double cellfun but... see a)

[Birdman]

Actually this approach worked for me.

[Birdman]

Oh, it says multiple cell arrays. Then,

 idx=cellfun(@(x) isinf(x{:}),meancalc)

this one works.

[Guillaume]

isinf(x{:})

will return Error using isinf. Too many input arguments if the cell array x contains more than one element. e,g.

x = {1 2 3}
isinf(x{:})

x{:} is expanded in the comma-separated list 1,2,3, so the above is equivalent to

isinf(1, 2, 3)

[Birdman]

Let me share the code that I worked so that we can clear the confusion:

%demo data
A(:,:,1)=[1 2;3 Inf];
A(:,:,2)=[Inf 2;5 6];
A(:,:,3)=[3 Inf;Inf 3];
meancalc=num2cell(A);
% what I understood by multilevel cell was each element was contained as a cell
meancalc=num2cell(meancalc)
%this found the indexes of Inf's
idx=cellfun(@(x) isinf(x{:}),meancalc)

[Guillaume]

I don't think that's what Sukry has (and a cell array where each element is a scalar would be a big waste of memory). A representative example would be

A{1} = {[1 2; 3 Inf], [1 2 3]; [Inf; -1; NaN], [3 Inf; Inf 3]}
A{2} = {[], [Inf 2; 5 6],  magic(3)}

That is you have matrices inside cell arrays themselves inside a cell array. This would be consistent with the notation A{i}{j}