How can I combine fields for two different structures in two different .mat files?

Hello, I am trying to combine two structures saved in two different .mat files into one structure/.mat file. I would like to end up with a new 1x1 structure containing the same fields and all the values within the fields for both original files.

My two files contain the same six fields (x, y, z, a, b, c) containing either cell or double (x,y,z are cell and a,b,c are double) with different dimensions.

I think I need to combine the fields and then create another structure from them, but I cannot figure out how to do this.

I think I may need to combine some vertically and some horizontally (1 x value vs. value x 1 dimension), which may be an issue.

Appreciate any help Thank you, Megan

[Megan Ladds]

I attached screenshots of the fields within my 1x1 structures as well as what I think I want the final result to be. I keep running into issues combining them since they are different dimensions, and I end up losing or combing things strangely. Your answers have gotten me closer than I was before and I'm hopeful it is possible.

Thank you for your help with this! Megan

[Walter Roberson]

Your train is 13821 x 228 in the first one and 3411x231 in the second one. That cannot be combined in one array 17232 x 459 without padding with some value, and it is not clear how you would like the data arranged. If the two datasets are T1 and T2, do you want blkdiag(T1,T2) ? That would be the right size, but would pad after T1 with 0's as columns, and would pad before T2 with 0's as columns.

Your logic appears to be that vectors should be converted to common orientation and concatenated, but that non-vectors should be blkdiag() together ? Would that be the case even if the non-vectors have a common dimension?

I would suggest to you that your logic is getting complex enough that it would be easier (and probably faster) to handle the fields by name with the logic for that field. The logic to detect which situation is being processed is not going to allow you to vectorize without the use of an auxiliary true function, at which point you might as well just give up and code more directly.

[Megan Ladds]

Alrighty, I figured it would come to something like that. Thank you so much for your help, I'll likely use the new info I have.

Answer 1

[Matt J]

Let s1 and s2 be the two structures. Then you can just do

 f=fieldnames(s1);
 for i=1:numel(f)

    scombined.(f{i})={s1.(f{i}), s2.(f{i})};

 end

[Megan Ladds]

This almost worked for what I needed (created a 1x1 structure), but I also need the values to be combined into one, rather than just creating 1x2 cells for each field with each of those cells containing the cells from both the structures' fields' in it. What would I need to add to your above suggestion to create that? Is that even possible?

Thank you!

[Matt J]

You can concatenate any way you like, e.g.,

r=@(z) reshape(z,1,[]);
scombined.(f{i})=[ r( s1.(f{i}) ) ,  r( s2.(f{i}) ) ];

Naturally of course, the field data must be shaped compatibly for concatenation.

[Walter Roberson]

You could probably do this more automatically by using struct2cell(), concatenating the two resulting cells, and cell2struct() back again.

If there is any possibility that the fields might not be in exactly the same order, you should use orderfields()

[Matt J]

I don't think so. Megan has said he contents of the cells have to be merged.

[Walter Roberson]

As I said, "concatenating the two resulting cells".

See my solution.

[Matt J]

Well, yes, but it's not "more automatic", right? You're still using an embedded for-loop via arrayfun.

Answer 2

[Walter Roberson]

A = struct('x', 1:2, 'y', 3:7);
B = struct('x',8:10,'y', 19:22);
temp = [struct2cell(A), struct2cell(B)];
result = cell2struct(arrayfun(@(ROWIDX) [temp{ROWIDX,:}], (1:size(temp,1)).', 'uniform', 0),fieldnames(A));

[Walter Roberson]

A more compact version with no temporary variable:

result = cell2struct(cellfun(@(P,Q) [P,Q], struct2cell(A), struct2cell(B), 'uniform', 0), fieldnames(A));

[Megan Ladds]

So would there be a way to do this and concatenate the cells if they aren't the same dimensions? The different dimensions within the fields (even though the fields are the same) is the root of my issue I believe.

[Walter Roberson]

Are all of the items vectors?

result = cell2struct(cellfun(@(P,Q) [P(:);Q(:)], struct2cell(A), struct2cell(B), 'uniform', 0), fieldnames(A));

[Walter Roberson]

Unfortunately I am away from my computer for a bit and will not be able to check until later.