Author: Robert Hyatt
Date: 13:39:15 09/14/98
Go up one level in this thread
On September 14, 1998 at 09:08:13, Jari Huikari wrote: >One my friend explained so well that even I understood the principle >of pruning search tree in chess program: > >I have found value for a move, by taking look for opponents possible >replies to it. When I evaluate next move, it's enough to find the first >opponents reply, which means worse score than that of previous move's. >And I don't have to take look to more replies, because I know previous >move was better. > >Implementing this however made my program play strange moves, sacrificing >queens for no reason and that kind of things. > >Now I ask, why that could be. Are there depths in iterative deepening, >which should be searched all the replies and exactly how bad reply >could be found, without doing a cutoff? > > Jari I'm going to assume this is not a "language problem". What you are trying to explain should be "alpha/beta". And it goes like this: After you search your first move at the root, and all of the opponent's replies, and all your replies to that and so forth, you get a score for the first move at the root. If your move ordering is good, the first move should be better than the second, correct? To confirm this, you search the second move, but again, if your move ordering is good, you quickly find a move by your opponent that "refutes" this move and you can stop as soon as you find a ply=2 move that produces a score that is worse for you than the score for the first move you searched. This is alpha/beta... For more details, there are lots of AI books around with it covered in detail... Bob
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.