On Mon, 21 Aug 2006 22:58:08 +0100, Matthew Toseland wrote: > Could it significantly slow down reaching optimality? Could it in fact be > partly responsible for the simulated O(n^2) swaps needed?
You mean swap attempts, of course, since the propability for a swap to happen at a given time has nothing to do with the total amount of swaps (as in swaps that actually occurred) needed, right ? > Anyway, should I change it to A/B, or keep it at A^2/B^2 ? > > On Mon, Aug 21, 2006 at 12:41:04AM +0200, Oskar Sandberg wrote: >> >> Yeah, the standard approach would be not the square anything. But it >> doesn't really matter: you are still optimizing in the right direction, >> it might just make it a little slower. >> >> // oskar >> >> > Here are the comments at the top of >> > freenet.node.LocationManager.shouldSwap(): >> > >> > * If A > B then they switch. >> > * >> > * If A <= B, then calculate p = A^2 / B^2. They then switch with >> > * probability p (that is, switch if rand.nextFloat() < p). >> > >> > I'm concerned about this last bit, if I recall the algorithm >> > correctly, you only square the A and B terms in the two dimensional >> > case, in 1 dimension it would just be p= A / B. >> > >> > Oskar, can you confirm or deny this? >> > >> > Ian. >> > >> > Ian Clarke: Co-Founder & Chief Scientist Revver, Inc. phone: >> > 323.871.2828 | personal blog - http://locut.us/blog >> > >> > >> > >> >> _______________________________________________ Devl mailing list >> Devl at freenetproject.org >> http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl >>
