Description |
To compute the Euclidean separation (L2 norm) between two sets of points in MATLAB can be slow and/or memory-hungry. In some cases (most particularly, if you are working with 2D, 3D or 4D data), this function will do it 2-4 times faster than the fastest m-code I've seen (due to Germano Gomes) and hundreds of times faster than a typical memory-efficient nested loop.
NB: for D much greater than 10-15, performance is better using GG's m-script. See the screenshot for a performance plot - green is GG, blue is mex_sepsq.
>> mex_sepsq_demo
A = randn(4, 5000);
B = randn(4, 5000);
C1 = mex_sepsq(A, B);
Elapsed time is 0.201335 seconds.
C2 = sepsq_gg(A, B);
Elapsed time is 0.517755 seconds.
Relative time per implementation: 1.00 2.57 |