Author: Severi Salminen
Date: 12:48:01 10/05/00
Go up one level in this thread
On October 05, 2000 at 12:33:47, Ralf Elvsén wrote: >On October 05, 2000 at 11:48:46, Severi Salminen wrote: > >>Hi! >> >>I ordered a copy of Heinz's book in the library of my home town. A few question >>about his AB explanation: >> >>1.He says: "While fail-highs with result>=beta place lower bounds on the real >>scores of failed searches, fail-lows with result<=alpha place upper bounds on >>them." >> >>So, what bound is he meaning (maybe my english is not so good): real scores of >>failed searches? I thought that Alpha and Beta are the only bounds involved - >>they are the lower and upper bounds of real score, aren't they. Does this imply >>in negamax implementation? I know basically what fail-high (and fail-low) means, >>but didn't understand this statement. >> > >Assume you are in a position with beta = (say) 100. >You search move A and get a score 150. Then you have a beta cutoff >and exit that node. But this value 150 is a lower bound on the "true" >score of that position because there might be moves you never serached >with values of 200 or whatever. So 150 is a lower limit on the "true" >value of the position. > >The reasoning is basically the same for the other case (if you think >about it for a while): where did all those values <= alpha come from? I understant the fail-high case. But maybe I don't know what is a fail-low: Could you check the next example, did I understand? In this exaple we make a 2 ply search and assume that black is in totally lost position. In root node we have two moves for white: exf6 and e6. Let's say that exf6 is the first move searched and after searching it we have A=900 (we captured a queen) and B=Inf. Then we search e6. After making that move we have A=-Inf and b=-900. Probably at first black responce we get a fail-high. So we return a 0 (material is even). And at root node we have a fail-low situation, right (A=900 and the score returned was 0)? So, this 0 should be an upper bound for e6 situation. Did I understand?? Severi "I was so clear" Salminen
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.