Any recommendation how to vectorize this double for loop?

I am a beginner in Matlab, and I would like to speed up the following:

X = 100;
MA = NaN(X,3);
for looper = 1:X
    compareDist = inf;
    array = MM(looper,:);
    for looper2 = 1:size(array,1)
        currPoint = cell2mat(C(looper,looper2));
        currDist = distancePoints3d(pP, currPoint);
        if(~isnan(currPoint(1))&&currDist<compareDist)
            compareDist = currDist;
            MA(looper,:) = currPoint;
        end
    end
end

Any help is appreciated. Thanks.

[per isakson]

What's &nbsp C ?

[UCL student]

C is a large cell containing points.

[per isakson]

Both &nbsp "cell" &nbsp and &nbsp "point" &nbsp are ambiguous. And &nbsp MM &nbsp what's that?

[UCL student]

MM is just used to measure the size. It should not be in the loop.

[per isakson]

"[MM] should not be in the loop." &nbsp However, in fact it is: &nbsp "array = MM(looper,:)".

Since &nbsp looper &nbsp is a scalar

    array = MM(looper,:);
    for looper2 = 1:size(array,1)

reduces to

    for looper2 = 1:1

??????

[Sean de Wolski]

How long does it take and how many times do you have to run it?

Answer 1

[lvn]

A couple of things:

1. If size(MM(looper,:),1) is independent of looper (as you suggest?), you can simply precalculate it before the loops.

2. Depending on your specific data cell2mat(C(looper,looper2)) can be vectorized by precalculating C2=cell2mat(C); be careful with indexing afterwards though (because all will be merged in one large array). Much easier, if C(looper,looper2) all have the same dimension is to get rid of cells altogether and store your C data in higher dimension matrices.

3. Test for ~isnan(currPoint(1)) before calling distancePoints3d, since if it is NaN the output of distancePoints3d is not used.

4. If you have managed to speed up the code, please post the new code, so we can work on the remainder :)