How to extract a matrix of values from cell array of cell arrays of structs

I have an array of array of structs, and I'd like to extract the value 'metric' from each, into a simple matrix. cell2mat and other functions just gave me various errors, so I wrote the following give-up code. It works, but is there a better way?

msea = zeros(n1,n2);
for i=1:n1 
    for j=1:n2 
        msea(i,j)=mse{i}{j}.metric; 
    end 
end

[Fangjun Jiang]

better to provide and show your test data

[Bruno Luong]

Your data is not array of array of structs but cell array of cell array of structs. That's are not convenient to organize the data.

You might use for example 2D struct array if they have the same fields.

[Jerry Guern]

@Bruno Luong Yes, it would indeed be easier if the data came already in form the I wanted it. Unfortunately, there's a module that outputs whole populated structs that I can't change. And we can't know in advance how large the 2D output will be in either dimension. So I'm stuck with the form it arrives in. But I have edited my question to say "...cell array of cell arrays..." Thanks for the clarification.

Answer 1

[Bruno Luong]

Generate data (Thanks Voss)

n1 = 3;
n2 = 4;
mse = cell(1,n1);
for ii = 1:n1
    mse{ii} = cell(1,n2); % row vector
    for jj = 1:n2
        mse{ii}{jj} = struct('metric',ii+(jj-1)*n1);
    end
end

Single line code

reshape([cell2mat(cat(1,mse{:})).metric],length(mse),[])

ans = 3×4

1 4 7 10 2 5 8 11 3 6 9 12

[Bruno Luong]

I would first convert to struct array S

S = cell2mat(cat(1,mse{:}));

Then use S from now on and forget about mse.

metric_array = reshape([S.metric], size(S))

[Jerry Guern]

Thank you! I especially appreciate the way you gave me what I actually wanted AND a more practical solution that I should have wanted.

Answer 2

[Walter Roberson]

msea = zeros(n1,n2);
for i=1:n1  
   msea(i,1:n2) = [mse{i}{1:n2}.metric]; 
end

[Jerry Guern]

@Walter Roberson Thanks! That's definitely an improvement over mine. But I'm trying to be disciplined about using vectors instead of loops, so the MATLAB gurus won't chastise me. :-)

[Stephen23]

"But I'm trying to be disciplined about using vectors instead of loops, so the MATLAB gurus won't chastise me"

This is a persistent myth about MATLAB, that loops should be avoided. Well written loops are neat and efficient.

Why specifically do you want to avoid loops?

Walter Roberson's answer will not work because the 2nd cell indexing produces a comma-separated list, into which further (dot) indexing is not possible. But if the data were in a structure array (as Bruno Luong recommended here), then something similar would be possible.

[Jerry Guern]

@Stephen23 Thanks. I literally have no reason to avoid loops beyond appeasing the "coding standards" monitors. Vectorizing does also make for tidier code in other languages, but this is not always so for matlab, like this case for example.

[Jerry Guern]

If I wasn't required (against my wishes) to vectorize my code, I'd do is this way. This is succinct and highly readable code. The two best Answers I got on how fully vectorize this were respectively LONGER than this method and a single line of deep matlab code that I needed the manual to decipher. Left to my own devices THIS is how I'd code this up. Thanks.