Author: José Carlos
Date: 03:15:35 04/21/01
Go up one level in this thread
On April 21, 2001 at 04:02:38, Steffen Jakob wrote: >On April 20, 2001 at 15:58:46, Dieter Buerssner wrote: > >>On April 20, 2001 at 15:34:01, Scott Gasch wrote: >> >>>I'm trying to come up with a good algorithm for detecting "easy" moves. The >>>goal is of course to make obvious recaptures faster. >> >>I am also in the process of thinking about this. Sorry, I cannot give an answer, >>just a few comments. >> >>>The obvious definition of an easy move is one where the score of the PV move at >>>a certain depth is delta better than all the other moves at the root position. >>>I use PVS, though, and because of the minimal window search on moves 2..N I >>>don't come up with exact scores for some moves. >> >>Even if you don't use PVS, the same problem exists. With fail hard alpha-beta, >>you will (almost?) allways only know, that the score of the remaining move is >>not better, than the score of the PV move. With fail soft alpha-beta, you may >>get better bounds, but in my experience, usually they will still be very close >>to alpha. >> >>>So I am looking for other definitions... right now I am experimenting with: if >>>the first move searched never changed from plys 1-7 and it recaptures on the >>>opponent's last move it is "easy". >> >>I think this can't work. Especially with hash tables, that are not cleared >>between moves, you will see very often, that the best move does not change in >>the first 7 plies, even when it would have changed, when starting a search from >>scratch without prefilled HTs. > >Hmmmm, in that case I don´t start at depth 1. If I have an exact value in the >hashtable I start the search with the depth of that entry. What do you do if the root moves have entries in the hash table with different depth, or any of them are not in the hash table at all? I think it's simpler to start a normal search, from depth=1. Then, as you'll hit the hash table until reaching a depth bigger than the stored one, you'll go really fast to that depth. This works fine for me in Averno. José C. >Greetings, >Steffen.
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.