Finding NaN values in Structure

Hi all,

I have the following code for finding a NaN value in a structure:

   structure.w.c = 5;
   structure.w.d = 4;
   structure.x.a = 1;
   structure.y.a = NaN;
   structure.z.a = 3;

   first_level = fieldnames(structure);
   structure_path_first = strcat('structure.',first_level);
   structure_path_complete = {};
   for i=1:1:length(first_level)
       temp = eval(structure_path_first{i});
       second_level{i} = fieldnames(temp);
       if strcmp(second_level{1,i},'a')
           structure_path_complete{i} = strcat(structure_path_first{i},'.',second_level{i});
           temp_2 = eval(char(structure_path_complete{i}));
           if isnan(temp_2) 
               value_nan{i} = structure_path_complete{i}; 
           end
       end
   end

The problem of the structure is that I do not know the names of the second level (w,x,y,z) and also not the number of array fields (a,d,c) in advance.

The code is working fine and I get the complete names of the array fields containing a NaN. The question now is, is there a smarter way than using for-loops finding these fields?

Thank you for your answers.

Cheers

Christian

[Stephen23]

Avoid eval, which makes code slow and buggy:

https://www.mathworks.com/matlabcentral/answers/304528-tutorial-why-variables-should-not-be-named-dynamically-eval

Answer 1

[Guillaume]

As others have said, do not use eval, particularly as there's a much easier way to convert a variable into a field name.

The way I would approach this is with a recursive function, avoiding hardcoding multiple loops. If the field being tested is a structure call yourself again:

function fnames = findnanfields(s)
   %s: a scalar structure
   %fnames: column cell array of field names that are nan. recurses through structures
   fnames = {};
   for fn = fieldnames(s)'
      fieldcontent = s.(fn{1});
      if isstruct(fieldcontent)
           subfields = findnanfields(fieldcontent);
           if ~isempty(subfields)
              fnames = [fnames; strcat(fn{1}, '.', subfields)];
           end
      elseif isnan(fieldcontent)
           fnames = [fnames; fn{1}];
      end
   end
end

This works regardless of the number of levels, number of fields, etc. However, it assumes that each field is scalar (as your code did).

Note that instead of returning the field hierarchy as a dotted string it may be better to return it as a cell array, depending on what you're planning to do with that information. In which case, replace the name generation by:

              fnames = cellfun(@(sn) [fn{1}, sn], subfields, 'UniformOutput', false);
              %instead of fnames = [fnames; strcat(fn{1}, '.', subfields)];

[Christian Baetz]

Thanks for this answer. The function is working fine. I assumed MATLAB provides a function to find such NaN-fields.

Answer 2

[Alexandra Harkai]

It is good practice to avoid eval:

first_level = fieldnames(structure);
counter = 0; % keep track of the found NaNs so far
for i = first_level' % row vector of field names
  temp = structure.(char(i));
  second_level = fieldnames(temp);
  for j = second_level' % row vector of field names
    if isnan( structure.(char(i)).(char(j)) )
      counter = counter + 1;
      value_nan{counter} = [ 'structure.', char(i), '.', char(j) ]; 
    end
  end
end

Alternatively, you can use arrayfun to loop through the field names:

value_nan = {}; % init as empty
function addPath(first, second) % function to register the path
  if isnan(structure.(char(first)).(char(second)))
      value_nan = [value_nan; ['structure.', char(first), '.', char(second)]]; % append new line
  end
end
arrayfun(@(first) arrayfun(@(second) addPath(first, second), fieldnames(structure.(char(first)))'), fieldnames(structure)'); % run through all first and second fields