Automatically creating an array of variables and assigning to a function with changing number of outputs

I am writing a script that creates an array (A) of vectors. The value of each vector is the set subscripts that describe it's position within the array. eg, for a three dimensional square array with side length 4:

 A{1} = [1,1,1]
 A{4} = [4,1,1]
 A{9} = [1,1,3]

Each vector is then passed to another function which calculates my final result. eg,

 results(i)=myfunction(A(i))

Im trying to automate the first stage, ie, assigning the value of the vector A{i} to be its subscript place within A. This will be variable on both the number of dimensions that A has and the side length (the sides are always the same length).

The method I have devised involves using ind2sub(). The number of outputs from in2sub is dependent on the dimensionality of the input matrix. To overcome this I am creating a character array with a variable for each dimension of A. eg for 3 dimensions.

 vector_name='[x,y,z]';

and then concatenate it with the string '=ind2sub(size(results),i)' :

 command_sting=strcat(vector_name,'ind2sub(size(results),i)')

so for my 3 dimension example command_string would store the character array:

 command_string=
                '[x,y,z]=ind2sub(size(results),i)'

I am then trying to use eval() to evaluate this statement but I get an error message:

 eval(command_string)
  Index exceeds matrix dimensions.

However if I type what is stored in command_string straight into the command line I do get the right result:

 i=1
 [x,y,z]=ind2sub(size(results),i)

 x =
     1
 y = 
     1
etc.

Why is the eval(command _string) not working? If anyone has a more elegant solution than this use of eval() I'd be grateful to hear it.

[Stephen23]

Evil eval strikes again!

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

Summary: using eval does not solve anything, it just makes complicated code that is impossible to debug.

[Stephen23]

"a more elegant solution"

Copy ind2sub and change it to output a vector. It is quite a simple change.

[Jan]

See FEX: augmented sub2ind .

Answer 1

[Guillaume]

You don't need to use eval to cope with varying number of outputs. Use cell arrays and their implicit conversion to comma-separated lists:

indices = cell(1, ndims(results));  %create a cell array the right size
[indices{:}] = ind2sub(size(results), someindex);  %assign c-s-l output to cell array
indexvector = [indices{:}];  %convert cell array to vector.

[Bob Hickish]

Thanks! Just what I was hoping for!

[Stephen23]

Note that ind2sub is a very simple function, and it is trivial to copy it and make your own version that returns a vector instead of separate output arguments (do NOT overwrite the original file).

Then you could avoid the whole bother altogether.

[Jan]

See FEX: augmented sub2ind .

Answer 2

[Adam]

[x, y, z] = ind2sub( [3,4], 9 );

works even though the size of the input is 2d, you just get z = 1. Likewise if you use n outputs you will just get 1 for those beyond the dimensionality of the input.

I would prefer this as an implementation personally, using the maximum dimensionality you will have. Possibly it may be better putting the results directly into an array rather than n named variables, but wither works.

Your code that comes after your eval would still have the problem of having to know what variables it needs to put into the A cell array so you might as well have all the variables there and ignore the ones not relevant for the given dimensionality.

[Bob Hickish]

Thanks for your answer. This would solve my problem, however I think the other solution above is a bit more elegant as it is generic without having unused variables hanging around.

[Adam]

Yes, I would agree. I've never tended to use comma-separated lists for these kind of things so I am always unfamiliar with where they can be used for assignments.