Posted: Apr 15, 2012 11:13 am
DanDare wrote:
No randomness is not optional. To systematically apply variations would require that you already know the optimal design. The only alternative to random variation is a brute force run through every possible combination. Combinatorial theory shows that the universe may not be around long enough to do that for even moderately complex systems.
Yes, it is, because the only difference between systematic coarse scans and random ones is the exact size of the interval. Like campermon, you seem to forget that combination of a coarse systematic scan be followed by a fine scan around the best regions, without creating a massive combinatorial problem. Using random intervals is like gambling, because you're hoping for a "lucky strike".
Having said that, this particular subject is a bit of a detour in this thread, so there is little point in pursuing it here.