Author: Miguel A. Ballicora
Date: 09:44:32 01/15/02
Go up one level in this thread
On January 15, 2002 at 12:13:06, Daniel Clausen wrote: >Hi > >On January 15, 2002 at 11:15:42, Miguel A. Ballicora wrote: > >>For me, it was significantly smaller than doing the same >>ordering I do in the rest of the tree. > >In other words, the root-like move-ordering scheme is >better than the move-ordering scheme in the rest of the >tree. root-like move-ordering scheme is better for the root but not necessarily for the tree. In the root, we have a stable nice position where both player have been playing reasonable good moves. In the tree, we have huge mess with ridiculous moves of all kinds. In that context, captures always have good chance to be a refutation. In the root not necessarily. Besides, this scheme can be only applied to nodes where you examine all the moves. That is, ALL nodes (that you cannot get the score above alpha) and PV nodes. It is really a mess to try to implement it somewhere else. I thought about it and I am going to try it eventually for the first few plies in the PV. What I did try was the following: count the nodes that need a certain move to be search in the tree. If I have to return alpha or below, that means that I have no "bestmove" by score but all the moves have been searched and I have the move that took "longer" to be searched. I stored this as bestmove in the hashtable so it can be used later. The tree was smaller in the first positin I tried, very exciting, but it was bigger in all the positions I tried later!!! all of them! What I did later was to store this move as "best" as long as it was searched more than 4 plies, so to improve the accuracy. I got better but still I did not get a smaller tree. I discarded the idea temporarily, it might be good in other situations but I doubt it. However, I feel that this kind of ordering in the PV could be benefitial just a little bit. Still, it is a mess when the PV change. The cleanest way would be to do a search function for the PV separated from search and keep a huge table to store all this. Maybe I can try only at the next ply from the root. I will try to implement something at one point and try it. Regards, Miguel Did you try to do root-like move-ordering in the rest >of the tree? Maybe for the first N plies, since we have to >store some statistic data for this. > >Sargon
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.