Author: John Coffey
Date: 10:53:31 09/28/98
My understanding of alpha-beta is like this ... 7 5 3 2 ? <--- some responses for black \ | / \ | 3 ? <--- some choices for white \ / \ / ? <--- current position Having looked at the first choice and all of black's responses, we conclude that the first move has a value of 3. We then try to determine the value of the second move. We see that one of the black responses has a value of 2, which is worse for white that the other choice of 3, so we already know that the second branch is going to be worse than the first, so we do not need to look at any more of the black responses, thus producing a savings. I have some questions .... 1. People are always talking about two values here, alpha and beta, but seems like we could be just dealing with one value, the value that we are comparing from the previous level. As a result I get confused when people talk about "fail low" and "fail high." 2. If doing a deeper search, then it seems to me that I would want to pass the same value even deeper into the tree .... 2 ? ? <----- Black responses at depth=4 \ |/ ? ? <---- white responses to black responses \ / 7 5 3 ? ? <--- some responses for black \ | / \ / 3 ? <--- some choices for white \ / \ / ? <--- current position In this case do we not look at the remaining black responses at depth =4 beceuse the best white move branch at depth=1 is a 3? I have found some pseudo-code on the net. I will get that a closer look. Thanks, John Coffey
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.