Better ways to achieve maintainable code involving n-d arrays?

Dear All,

I have a question on the best way to write maintainable and readable code inolving n-d arrays.

To explain my problem, look at the following code snippet involving "someArray" which is a m-by-ntime matrix,

the second dimension is time say.

I initialise the array and then step forward in time using an evolution equation:

someArray (:,1) = intialisation(someValues);
for tstep = 2:ntime
 someArray (:,tstep) = evolution_equation( someArray (:,tstep - 1), other_stuff );
end

This innocent piece of code is a maintenance nightmare: Imagine in a next revision someArray would have a

different dimension say m-by-ntime-by-k-by-j. Even if the logic of the calculation does not change a bit I have

to modify every expression involving someArray for purely syntactical reasons:

someArray (:,1,:,:) = intialisation(someValues);
for tstep = 2:ntime
 someArray (:,tstep,:,:) = evolution_equation( someArray (:,tstep - 1,:,:), other_stuff );
end

Needless to say I have plenty of "someArrays" and many lines of code like this and I get sick and tired of

doing those "mindless" modifications.

So my question is simply:

• Do you encounter this problem as well?
• Do you know of a better (= more maintenance friendly) way of doing this?

Thank you gg

Answer 1

[Oleg Komarov]

You may find useful the comma-separated list expansion:

clear B
tstep = 1;
dims = [2,2,2,2];
A = rand(dims);

% Comma-Separated List
csl = cellstr(repmat(':',ndims(A)-2,1));

% Dynamic assignment with csl expansion 
B(:,tstep,csl{:}) = sum(A,2);

Try to change the value of dims to see what happens.

Oleg

[Jan]

Typo: "csle"->"csl". I'd prefer this to create the cell string: csl = cell(1, ndims(A)-2); csl(:)={':'};

But for the small number of expected dimensions the absolute speed difference is negligible.

[gg]

hmm, if you always append to the right this might work. But I consider this more a partial workaround than a solution.