Author: David Rasmussen
Date: 10:31:12 01/15/02
Go up one level in this thread
On January 15, 2002 at 11:15:42, Miguel A. Ballicora wrote: >On January 15, 2002 at 08:51:26, David Rasmussen wrote: > >>On January 15, 2002 at 08:45:21, Severi Salminen wrote: >> >>>>Many people use node count (or time) to order root nodes, apart from always >>>>ordering the pv move of last ply first, of course. This seem to work ok. But as >>>>some people have suggested (Christophe, for one), this method isn't very much to >>>>the point and might have unwanted sideeffects. >>> >>>How is it "not to the point" and what side effects he means? For me it works. >>> >>>Severi >> >>It works for me too. >> >>The phrases "not to the point" and "side effects" are mine, not Christophe's. >>But he has expressed similar things. > >It works for me and I do not care if it is indirect or whatever. Because >there is a real simple way to see it is working. Measure the tree size >and that's it. For me, it was significantly smaller than doing >the same ordering I do in the rest of the tree. It had another positive side >effect! Killers started to work! Killer did not work for me, it was a wash. >When I added the root ordering, Killers added a ~15% decrease in tree size ( >as far as I remember). > >Now the question is... is there a better way to do it? I am all ears :-) > >>When I said "not to the point", I meant that 'nodes used' is not a very direct >>measure of the quality of a move, it's rather an indirect one. If I somehow >>could get the score of each root move after an iteration, I would certainly use >>that for ordering, but I can't since I use PVS. > >I am not sure if I agree. I do not want ordering accoring to the score, what >I want is "what is the move that has more chances to beat the main move?" >Sometimes, it is not the second best move (according to the evaluation after >depth = X) but another move that is "promising" even though at depth X does not >have a high score but it has chances to have a high one at score X + 1. For >instance, attacking moves that fall just short. > >Regards, >Miguel > > >> >>I can't name any tangible side effects, I said it "might" have side effects. >> >>/David Good points. I tend to agree with you. I am just not sure that the final truth has been found about root move ordering. Especially when such a strong program as Tiger does _not_ use node ordering. /David
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.