Author: Russell Reagan
Date: 16:14:18 07/02/04
Go up one level in this thread
On July 02, 2004 at 15:47:25, Scott Gasch wrote: >Hashing mate scores is tricky. The thing to realize is that when you have a >mate in n score at a node it's relative to the root of the search. So if you >have mate in 8 ply (from the root) at some node that is 3 ply deep then what you >really have is mate in 5 ply (from here). That is the number you want to hash. >Then when you pull it out of the hash you can readjust it... Say you pull it >out of the hash at a node that is 6 ply deep in the search. You say "ah, this >move is mate in 5 ply from here, and I am 6 ply deep, so it's real score is mate >in 11 ply (from the root)". Wouldn't there still be a problem from searches that start at different points in the game? For instance, at move 10 of the game you do your search. At depth 3 there is a mate in 6, which is a mate in 9 from the root. On your next move you do your search, and at depth 5 you hit the same position and return "mate in 6 from here", which is a mate in 11 from the root, which isn't the right score. >Or you can do what Bruce talks about which is when you store a mate-in-n exact >score in the hash just convert it to a upper or lower bound of mate-in-100 and >forget about it. Okay, so the problem is that we typically deal with relative mate scores. We store "mate in N ply from here". The problem is "here" will be different all over the tree and at different points in the game. Instead, what if we deal with "mate at ply N" scores? I think that should work, but it seems a little too simple. Surely I'm not the first to think of this. So what am I overlooking?
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.