Is there an alternative to eval in a struct?

Hi,

I know that eval is evil, but in this particular case, I couldn't find any other way to do it, so I wanted to ask you if anyone comes up with a more efficient way.

The problem is, I need to define a struct, and structs inside that struct, and more structs inside that struct. I think it would be better to describe it (simplified) this way:

subject = {'subj_1','subj_2','subj_3'};
runs = {'rest','task1','task2','task3'};
roigroups = {'a','b','c'};
a = {'dlpfc','amyg'};
b = {'insula','postcg'};
c = {'thalamus','nucl_ruber'};

subjects = struct();
% The dirty way, assume that other structs defined previously:
for i = 1:length(subject)
    for p = 1:length(runs)
        for x = 1:length(roigroups)
            for y = 1:eval(strcat('length(',roigroups{x},')'))
                subjects.(subject{i}).(runs{p}).(roigroups{x}).(eval(strcat(eval('roigroups{x}'),'{y}'))) = struct();
            end
        end
    end
end

Unrecognized function or variable 'tasks'.

It works really well, manually checked, but it's a pain to write and read. Is there a workaround for this type of task?

[Akira Agata]

The code doesn't work correctly. Could you tell us what the desired output looks like?

[Matt J]

Should this line be,

%runs = {'rest','task1','task2','task3'};
tasks = {'rest','task1','task2','task3'};

[Akira Agata]

Still not working.

As shown below, error occurs in the line "subjects.(subjects{i})~" .

subjects = {'subj_1','subj_2','subj_3'};
tasks = {'rest','task1','task2','task3'};
roigroups = {'a','b','c'};
a = {'dlpfc','amyg'};
b = {'insula','postcg'};
c = {'thalamus','nucl_ruber'};
% The dirty way, assume that other structs defined previously:
for i = 1:length(subjects)
    for p = 1:length(tasks)
        for x = 1:length(roigroups)
            for y = 1:eval(strcat('length(',roigroups{x},')'))
                subjects.(subjects{i}).(tasks{p}).(roigroups{x}).(eval(strcat(eval('roigroups{x}'),'{y}'))) = struct();
            end
        end
    end
end

Unable to perform assignment because dot indexing is not supported for variables of this type.

[Yasir Çatal]

Sorry, I edited now so that it works. Should've checked before, sorry about that.

Answer 1

[Matt J]

Your data organization seems like it should be reconsidered. In a struct, it doesn't make sense if different fields have essentially the same kind of contents. What I think you really want is a 2D or 3D struct array, something like

subject = {'subj_1','subj_2','subj_3'};
tasks = {'rest','task1','task2','task3'};

M=numel(subject);
N=numel(tasks);

a = struct('dlpfc',[],'amyg',[]);
b = struct('insula',[],'postcg',[]);
c = struct('thalamus',[],'nucl_ruber',[]);

[Subjects(1:M,1:N).a]=deal(a);
[Subjects(1:M,1:N).b]=deal(b);
[Subjects(1:M,1:N).c]=deal(c)

Subjects = 3×4 struct array with fields:
    a
    b
    c

[Yasir Çatal]

Okay, but is there an intuitive way to see which task/roi etc. belong to which subject? There are some subject specific things, for example, I need to exclude some subjects, and when I do that, I don't want every subsequent subject to go up one row and mess everything.

[Stephen23]

@Yasir Çatal: sure, but Matt J's suggestion is a much better use of MATLAB to achieve that. You can still use the meta-data in roigroups to select/filter/whatever by defining indices. That would a simple, versatile, efficient MATLAB approach. Akira Agata's approach is also very good, and would make processing your data much simpler because you could use the standard set of table analyzing/processing tools.

[Matt J]

@Yasir Çatal The essence of my proposal is that you currently are setting up a nest of structures with 3-dimensional dot indexing. A typical lookup will look like,

Subjects.subj1.task.c

Here sub1 and task2 could easily just be integer indices. There's no apparent reason they have to be field names. With integer indexing, the data is much easier and faster to access

Subjects(1,2).c

and is also much easier to initialize as my answer illustrates.

[Matt J]

Okay, but is there an intuitive way to see which task/roi etc. belong to which subject?

In your post, all subjects share the same tasks and roigroups. That's why I think an array makes more sense.

for example, I need to exclude some subjects, and when I do that, I don't want every subsequent subject to go up one row and mess everything.

You would have to show an example of "mess everything". Why couldn't you extract a subset of Subjects using integer indices? e.g.,

subset=Subjects([1,5,7],:)

[Yasir Çatal]

Thanks to everyone that involved, I guess I should go with subsets.

Answer 2

[Akira Agata]

I believe it's better to arrange the data as a table rather than the deeply nested structure. How about the following?

subject = {'subj_1','subj_2','subj_3'};
runs = {'rest','task1','task2','task3'};
roigroups = {'a','b','c'};
a = {'dlpfc','amyg'};
b = {'insula','postcg'};
c = {'thalamus','nucl_ruber'};

[p,q,r] = ndgrid(subject, runs, [a,b,c]);

T = table(p(:),q(:),r(:),'VariableNames',{'subject','runs','roi'});
T.roigroups = repelem(roigroups',24,1);
T = movevars(T,'roigroups','Before','roi');

disp(T)

     subject        runs       roigroups         roi      
    __________    _________    _________    ______________

    {'subj_1'}    {'rest' }      {'a'}      {'dlpfc'     }
    {'subj_2'}    {'rest' }      {'a'}      {'dlpfc'     }
    {'subj_3'}    {'rest' }      {'a'}      {'dlpfc'     }
    {'subj_1'}    {'task1'}      {'a'}      {'dlpfc'     }
    {'subj_2'}    {'task1'}      {'a'}      {'dlpfc'     }
    {'subj_3'}    {'task1'}      {'a'}      {'dlpfc'     }
    {'subj_1'}    {'task2'}      {'a'}      {'dlpfc'     }
    {'subj_2'}    {'task2'}      {'a'}      {'dlpfc'     }
    {'subj_3'}    {'task2'}      {'a'}      {'dlpfc'     }
    {'subj_1'}    {'task3'}      {'a'}      {'dlpfc'     }
    {'subj_2'}    {'task3'}      {'a'}      {'dlpfc'     }
    {'subj_3'}    {'task3'}      {'a'}      {'dlpfc'     }
    {'subj_1'}    {'rest' }      {'a'}      {'amyg'      }
    {'subj_2'}    {'rest' }      {'a'}      {'amyg'      }
    {'subj_3'}    {'rest' }      {'a'}      {'amyg'      }
    {'subj_1'}    {'task1'}      {'a'}      {'amyg'      }
    {'subj_2'}    {'task1'}      {'a'}      {'amyg'      }
    {'subj_3'}    {'task1'}      {'a'}      {'amyg'      }
    {'subj_1'}    {'task2'}      {'a'}      {'amyg'      }
    {'subj_2'}    {'task2'}      {'a'}      {'amyg'      }
    {'subj_3'}    {'task2'}      {'a'}      {'amyg'      }
    {'subj_1'}    {'task3'}      {'a'}      {'amyg'      }
    {'subj_2'}    {'task3'}      {'a'}      {'amyg'      }
    {'subj_3'}    {'task3'}      {'a'}      {'amyg'      }
    {'subj_1'}    {'rest' }      {'b'}      {'insula'    }
    {'subj_2'}    {'rest' }      {'b'}      {'insula'    }
    {'subj_3'}    {'rest' }      {'b'}      {'insula'    }
    {'subj_1'}    {'task1'}      {'b'}      {'insula'    }
    {'subj_2'}    {'task1'}      {'b'}      {'insula'    }
    {'subj_3'}    {'task1'}      {'b'}      {'insula'    }
    {'subj_1'}    {'task2'}      {'b'}      {'insula'    }
    {'subj_2'}    {'task2'}      {'b'}      {'insula'    }
    {'subj_3'}    {'task2'}      {'b'}      {'insula'    }
    {'subj_1'}    {'task3'}      {'b'}      {'insula'    }
    {'subj_2'}    {'task3'}      {'b'}      {'insula'    }
    {'subj_3'}    {'task3'}      {'b'}      {'insula'    }
    {'subj_1'}    {'rest' }      {'b'}      {'postcg'    }
    {'subj_2'}    {'rest' }      {'b'}      {'postcg'    }
    {'subj_3'}    {'rest' }      {'b'}      {'postcg'    }
    {'subj_1'}    {'task1'}      {'b'}      {'postcg'    }
    {'subj_2'}    {'task1'}      {'b'}      {'postcg'    }
    {'subj_3'}    {'task1'}      {'b'}      {'postcg'    }
    {'subj_1'}    {'task2'}      {'b'}      {'postcg'    }
    {'subj_2'}    {'task2'}      {'b'}      {'postcg'    }
    {'subj_3'}    {'task2'}      {'b'}      {'postcg'    }
    {'subj_1'}    {'task3'}      {'b'}      {'postcg'    }
    {'subj_2'}    {'task3'}      {'b'}      {'postcg'    }
    {'subj_3'}    {'task3'}      {'b'}      {'postcg'    }
    {'subj_1'}    {'rest' }      {'c'}      {'thalamus'  }
    {'subj_2'}    {'rest' }      {'c'}      {'thalamus'  }
    {'subj_3'}    {'rest' }      {'c'}      {'thalamus'  }
    {'subj_1'}    {'task1'}      {'c'}      {'thalamus'  }
    {'subj_2'}    {'task1'}      {'c'}      {'thalamus'  }
    {'subj_3'}    {'task1'}      {'c'}      {'thalamus'  }
    {'subj_1'}    {'task2'}      {'c'}      {'thalamus'  }
    {'subj_2'}    {'task2'}      {'c'}      {'thalamus'  }
    {'subj_3'}    {'task2'}      {'c'}      {'thalamus'  }
    {'subj_1'}    {'task3'}      {'c'}      {'thalamus'  }
    {'subj_2'}    {'task3'}      {'c'}      {'thalamus'  }
    {'subj_3'}    {'task3'}      {'c'}      {'thalamus'  }
    {'subj_1'}    {'rest' }      {'c'}      {'nucl_ruber'}
    {'subj_2'}    {'rest' }      {'c'}      {'nucl_ruber'}
    {'subj_3'}    {'rest' }      {'c'}      {'nucl_ruber'}
    {'subj_1'}    {'task1'}      {'c'}      {'nucl_ruber'}
    {'subj_2'}    {'task1'}      {'c'}      {'nucl_ruber'}
    {'subj_3'}    {'task1'}      {'c'}      {'nucl_ruber'}
    {'subj_1'}    {'task2'}      {'c'}      {'nucl_ruber'}
    {'subj_2'}    {'task2'}      {'c'}      {'nucl_ruber'}
    {'subj_3'}    {'task2'}      {'c'}      {'nucl_ruber'}
    {'subj_1'}    {'task3'}      {'c'}      {'nucl_ruber'}
    {'subj_2'}    {'task3'}      {'c'}      {'nucl_ruber'}
    {'subj_3'}    {'task3'}      {'c'}      {'nucl_ruber'}

Answer 3

[Steven Lord]

For most of what you're trying to do you could use setfield.

subject = {'subj_1','subj_2','subj_3'};
runs = {'rest','task1','task2','task3'};
roigroups = {'a','b','c'};
a = {'dlpfc','amyg'};
b = {'insula','postcg'};
c = {'thalamus','nucl_ruber'};

subjects = struct();
subjects = setfield(subjects, subject{1}, runs{2}, roigroups{3}, c{1})

subjects = struct with fields:
    subj_1: [1×1 struct]

% Checking
subjects.subj_1

ans = struct with fields:
    task1: [1×1 struct]

subjects.subj_1.task1

ans = struct with fields:
    c: 'thalamus'

For that last input probably I'd make a cell array of cell arrays out of a, b, and c.

D = {a, b, c};
subjects = setfield(subjects, subject{2}, runs{1}, roigroups{2}, D{2}{2})

subjects = struct with fields:
    subj_1: [1×1 struct]
    subj_2: [1×1 struct]

subjects.subj_2.rest.b

ans = 'postcg'