Calculate values of struct of struct

I have a struct with several fieldnames, each of them is again a struct and has a fieldname "P". This "P" is an array.

I want to calculate the sum of all "P "s like:

s.s1.P = (1:20).'; % For simplicity I have assigned dummy values to the "P "s.

s.s2.P = (1:20).';

s.s3.P = (1:20).';

% ...

result = s.s1.P + s.s2.P + s.s3.P % + ...

I have already a solution with a for loop:

field = fieldnames(s);

result = zeros(size(s.(field{1}).P));

for i = 1:length(field)

result = result +s.(field{i}).P;

end

I also tried to use "structfun" but were not able to get this result.

Is it possible to calculate this problem with "structfun" or another function or is the for loop the final solution for this problem?

[Stephen23]

Rather than forcing pseudo-indices into fieldnames you should simply use a structure array, as dpb showed.

Answer 1

[Bruno Luong]

Single line (work for fields with struct NOT objects of class, struct is NOT some sort of class)

s.s1.P = (1:20).';  % For simplicity I have assigned dummy values to the "P "s.
s.s2.P = (1:20).';
s.s3.P = (1:20).';

sum([cat(1,struct('sx',struct2cell(s)).sx).P],2)

ans = 20×1

3 6 9 12 15 18 21 24 27 30

Answer 2

[dpb]

The problem of using sequentially-named fields in the struct instead of an array of struct; there's no good way to process the resulting variables or struct fields programmatically when do so.

structfun applies the function to each field of a scalar struct so it isn't the tool for the purpose here, anyway, you're trying to process across fields, not within each. And, arrayfun is for an array but you didn't use an array to hold the similar data, you chose to use multiple fields instead. A 2D array would have been an alternate choice that could have handled easily...

s.P = (1:20).';
s.P = [s.P (1:20).'];
s.P = [s.P (1:20).'];

result = sum(s.P,2)

result = 20×1

3 6 9 12 15 18 21 24 27 30

Or, alternatively, as a struct array

s.P = (1:20).';
s(2).P = (1:20).';
s(3).P = (1:20).';

result = sum([s.P],2)

result = 20×1

3 6 9 12 15 18 21 24 27 30

[Julian]

Earlier I solved it with a cell.

I originally forgot to say that my inner struct is really a custom class, which is kind of like a struct.

The second variant is interesting for me and I will probably implement it, since I can hopefully save the cellfun in the later code flow.

[Bruno Luong]

A class? But you said in the question "I also tried to use "structfun" but were not able to get this result" How can you even possible use structfun on a class???

[Julian]

Sorry I am working for too long.

I meant an object, an instance of a class.

You can use "structfun" for structs (and for objects, because structs are special Matlab objects.The second part of this sentence is wrong)

[Bruno Luong]

"structs are special Matlab objects"

No, this is incorrect.

[Paul]

structfun can be used on class objects?

[Bruno Luong]

No

D=delaunayTriangulation(rand(10,2))

D = 
  delaunayTriangulation with properties:

              Points: [10×2 double]
    ConnectivityList: [12×3 double]
         Constraints: []

structfun(@(D)D.Points, D)

Error using structfun
Inputs to STRUCTFUN must be scalar structures.

Sorry my first remark on structfun on class is not applicable int the present context, since s is still a struct with 3 fields. structun can be applied on s.

[Julian]

Yes structfun was used on s, a struct and it's not posible to use structfun on class objects.

Answer 3

[Bruno Luong]

Honestly I prefer the for-loop

s.s1.P = (1:20).';  % For simplicity I have assigned dummy values to the "P "s.
s.s2.P = (1:20).';
s.s3.P = (1:20).';

c = struct2cell(structfun(@(sf) sf.P, s, 'UniformOutput', false));
sum(cat(2,c{:}),2)

ans = 20×1

3 6 9 12 15 18 21 24 27 30

[dpb]

Given the already bad idea of having created the sequentially-named fields, agree, the for loop is the way; as your example shows, to do anything else gets far too convoluted and undoubtedly would be slower besides.

At least with struct fields, one can programmatically access them w/o the eval abomination.

[Julian]

Thanks for the solution.

Unfortunately, I already came up with this, but I didn't like this huge nesting of functions either.

I understand why you would prefer the for-loop.