Question

Possible Duplicate:
Growable data structure in MATLAB

So in my current MATLAB script, I have a very large indeterminate-sized growing array. There is currently nothing I can do about it, because if I actually preallocate, it would take many many many times more memory than it should need (maximum possible amount of values is 640 per pixel, but usually it’s something along the lines of 2-5).

Usually in this case, I’d be using a vector in C++ or something, where it grows exponentially in relation to a given capacity. But I think the matrices in Matlab starts fragmenting much faster than the purpose driven C++ vectors.

What do you guys think is the best alternative to something like this? Or should I just stick with normal arrays and hope that adding around 100k elements sequentially will work?

Thanks in advance.

Answer 1

You could try what std::vector does when reallocating elements — double its capacity every time it fills up which has an amortized cost of O(1).

Test:

function test_MAT_vector

    LIM = 5e4;
    %%# Normal
    tic;
    A = zeros(1,1);
    for i = 1:LIM
        A(end+1) = i;
    end
    toc

    %%# std::vector clone
    tic;
    B = zeros(1,1);
    for i = 1:LIM
        if(i > numel(B))
            B = [B;zeros(numel(B),1)];
        end
        B(i) = i;
    end
    toc

end

Output:

Elapsed time is 3.489820 seconds.

Elapsed time is 0.006710 seconds.

Using cell

%%# cell
tic;
A = cell(1,1);
for i = 1:LIM
    A(end+1) = {i};
end
toc

Elapsed time is 2.740792 seconds.