Speed and Multiple Vectors

Is it faster to operate on multiple individual vectors or to put many vectors into a single matrix and simply reference the columns of the resulting matrix?

Answer 1

[per isakson]

This test probably has little meaning. (At least, Matlab is not smart enough to see that I take max of the same vector thousand times.) I guess, the difference is because the vector remains in a cache closer to the cpu. If speed is important, you might want to make a more realistic test.

    N = 1000;
    M = rand(1e5,N);
    v = M(:,1);
    tic
    for jj = 1 : N 
       m = max(M(:,jj)); 
    end
    toc
    tic
    for jj = 1 : N 
       m = max(v); 
    end
    toc

returns

    Elapsed time is 0.465941 seconds.
    Elapsed time is 0.036675 seconds.
.

One column only:

    N = 1000;
    M = rand(1e5,N);
    v = M(:,1);
    JJ  = floor(N/2);
    tic
    for jj = 1 : N 
       m = max( M( :, JJ ) ); % <<<<<<<<<<<
    end
    toc
    tic
    for jj = 1 : N 
       m = max(v); 
    end
    toc

returns

    Elapsed time is 0.340061 seconds.
    Elapsed time is 0.046989 seconds.

Answer 2

[Star Strider]

I suggest combining them into a single matrix and referencing the columns of the resulting matrix. It’s a lot easier to code and store those results. This works for vectors of equal lengths.

If you encounter the problem of the vectors having different lengths, it°s easy to use a cell array to store them and have them behave essentially as a matrix (except for not being able to use that matrix in matrix computations). It is easy to convert them back to numerical vectors for computation when you need to.