Author: David Mitchell
Date: 13:50:26 02/22/04
Go up one level in this thread
On February 22, 2004 at 15:27:44, Bob Durrett wrote: >In http://www.talkchess.com/forums/1/message.html?350711, Christophe Theron >made the fascinating and thought-provoking observation that: > > >"Conceptually, an alpha-beta search is doing several thousands takebacks per >second." > > >I do not doubt that this is true. However, I've never heard the alpha/beta >described quite that way. > >Could somebody please relate this to the more conventional concept of >alpha/beta? > >Bob D. Very simple, really. Before any position is evaluated, the move that leads to that position is "made", on a data struct inside the program. The internal "board", if you will. But is that position the best? The program can't tell, without comparing it to thousands, perhaps many thousands, of other positions. Each position is preceeded by the "move" that leads to it. Which has to be "made" and then "unmade", and then the next one "made", and "unmade", etc.. The important thing to understand A/B, to me, is that _first_ you make the moves to a depth you choose, and only _then_ (in only from a quiet position), is the position evaluated thoroughly and scored. Then, A/B takes those scores and the moves that made them, and works back _up_ toward the root position. Best moves, and scores are _lifted_ up from the tips, back toward the root, (starting position), and scores for your opponent's moves are negated (so if they're good for your opponent, they're bad for you), and vice-versa. All involving many moves being "made" and "unmade", every second. It's quite an amazing algorithim, actually. It's no wonder virtually every chess program uses it so extensively. dave
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.