Concatenate cell arrays from structs

I have the following struct named output.

Inside this output I have 27 fields with their values. Their values are 1x8 cell each. Each cell has different number of rows.

I want to:

Concatenate all the cell arrays (struct values) in such a way where the final result will be an array with all 8 columns and the rows of each cell array.

Here is what it looks like:

struct=output

field        value
smooth       1x8 cell array
rms          1x8 cell array
stft         1x8 cell array
haar         1x8 cell array
sym          1x8 cell array
.            .
.            .   
.            .
.            .

Each cell array has different number of rows!

Can I concatenate all the cell arrays into one? If yes, is there any efficient way to do this?

Thank you in advance!

[Stephen23]

"Each cell array has different number of rows!"

Not according to your description and the screenshot: each cell array has exactly one row: all of the cell arrays are size 1x8. The size of the arrays inside the cells is totally irrelevant.

Do not confuse the contents of the cell arrays with the cell arrays themselves: when those cells are concatenated into one cell array (as your question requests and the answers show how) then whatever data is inside them (and what size those arrays are) is still totally irrelevant. Are you expecting to also concatenate the contents of each cell along each column of the cell array?

[Ioannis Agalliadis]

Sorry for the misunderstanding. Lesson learned. Yes, I want to concatenate also the contents of each cell array. I used cell2mat but it overtook RAM and I couldn't do any other action on my computer!

[Guillaume]

It sounds like what you want is something different than what you appear to have asked. But you have accepted an answer so I'm not sure.

cell2mat will temporarily require twice as much memory as necessary. In your screenshot, the column vectors in a single cell array all have the same height. If that is indeed the case, then your fields should just be an Nx8 matrix rather than a 1x8 cell array of Mx1 vectors. This would be more memory and speed efficient.

[Ioannis Agalliadis]

But still the other fields have different number of rows.

Here is how I save the data into the struct:

for i=1:numel(featureNames)
                featureName = featureNames{i};
                for j=1:8
                    output.(featureName){j} = obj.features.(featureName).run(data(:,j));
                end
            end

So, instead of saving that way you recommend me to save it how?

[John BG]

Calimera Ioannis

This is John BG <mailto:[e-mail] [e-mail]>

Guillame and Stephen Cobeldick are very experienced MATLAB forum contributors, the kind that usually with a single shot they get the right answer.

But I'd like to point out the fact that unless you post the structure, or a significant chunk of it, you are keeping them, and all other readers, blind folded: Like in a Mexican pinyata party, guessing where the pinyata is.

So, please would you be so kind to make available the structure, as attachment, as .mat file, the whole structure, or a sample of it.

If containing sensitive data, just make a copy, and edit values with noise.

Regards

John BG

<mailto:[e-mail] [e-mail]>

[Ioannis Agalliadis]

I found out how to solve my issue.

This is how it inserts the contents of each matrix of a cell into one final matrix:

for i=1:numel(featureNames)
                featureName = featureNames{i};
                for j=1:numel(output.(featureName))
                    matrix(offset+1:offset+obj.dimensions.(featureName),j) = output.(featureName){j};
                end
                offset = offset + obj.dimensions.(featureName);
            end

With the variable offset I keep track of the number of rows of each matrix of each cell and then I concatenate them in a final matrix which dimensions are a large number of rows x with 8 columns. The sad thing is, that this was not the preferred matrix I should create but a different one.

Answer 1

[Guillaume]

1x8 cell array means 8 columns and one row. So it looks like your cell arrays have all the same number of rows, exactly one.

Perhaps you mean the contents of the cell arrays, each cell array has 8 something, but you haven't told us anything about what these somethings are (numeric matrices? column vectors maybe? char arrays?).

I'm very unclear on what you want to achieve. If you want to concatenate the cell arrays of all the fields together, then it's easily done:

c = struct2cell(youstructure);
c = vertcat(c{:});

This will create an mx8 cell array, where m is the number of fields. The size of the contents of the cell array have no influence on this.

[Ioannis Agalliadis]

Sorry for the insufficient information. Check the image I added! I want to concatenate the contents of each cell array which has an arbitrary number of rows but 8 columns. These cell arrays has double float numbers. The final product I want to be is a large array with all these rows and columns combined into one.

[Guillaume]

I'm still not clear on your data structure. But first, is the above solution what you're after or not all?

if your cell arrays are 1x8 then it is the cell arrays that have 8 columns (and one row).

This is a 1x8 cell array:

c1 = {[], [], [], [], [], [], [], []}

whose contents is all empty (0x0) double matrices. This is another 1x8 cell array

c2 = arrayfun(@magic, 1:8, 'UniformOutput', false)

whose contents are double matrices of varying sizes (1x1 to 8x8).

So what is the size of the cell arrays, and what is the size and type of the contents of each cell? Probably best, would be for you to upload a mat file with your structure so we can see for ourselves what's what.

[Ioannis Agalliadis]

In the updated question there is an image where you can see the dimensions of the cell arrays I am talking about.

Answer 2

[Stephen23]

You could use struct2cell:

C = struct2cell(output);
C = vertcat(C{:});

Note that the order of fields in a struct can change!

[Ioannis Agalliadis]

What do you mean that the order can change? Does it mean, that one time can save the fields in order A,B,C,D and another time can save them in C,D,A,B ?

[Guillaume]

The order of the fields is defined by the order in which they were created.

 s1 = struct('a', 1, 'b', 2)
 c1 = struct2cell(s1)

 s2 = struct('b', 2, 'a', 1)
 c2 = struct2cell(s2)

For all intent and purpose s1 and s2 are identical, but the result of the conversion to cell array is not

isequal(s1, s2)  %returns true. The structures are supposedly identical
isequal(c1, c2)  %returns false. The cell array are definitively not the same

The cell arrays are ordered in the order of the fields as returned by fieldnames.

[Stephen23]

@Ioannis Agalliadis: You do not state what you are using to create those fields, so how am I supposed to know if they could be created in a different order by whatever creates them? Once they are are created then the order could be changed via orderfields. Therefore any code that assumes that the fields are in a particular order is going to be fragile unless you order them yourself just before converting to cell. My point is that struct2cell does not tell you if the fields have changed order or if any fields are missing/extra, therefore you need to make sure that your code is robust to these possibilities.