Author: Bruce Moreland
Date: 13:41:00 01/20/99
Go up one level in this thread
On January 20, 1999 at 16:35:31, Bruce Moreland wrote: > >On January 19, 1999 at 20:20:48, Larry Griffiths wrote: > >>I ran my program with Alpha-Beta and without Alpha-beta. >>The scores and moves chosen were different. >> >>Is this normal for Alpha-Beta or is this an indication of an >>error in my implementation of Alpha-Beta pruning? >> >> >> >>Also, The time is 1052.58 seconds without Alpha-Beta >>and 3.3 seconds with Alpha-Beta. >> >>What ratio of elapsed time would indicate that I have achieved >>good move ordering etc. with the Alpha-Beta pruning? > >You should get the same score unless: > >1) You have a hash table that you use to prune, and do any path-dependent >scoring such as 50-move rule, or repetition detection. > >2) You do forward pruning based upon alpha or beta. This includes null move >forward pruning. > >3) You do any sort of imperfect hashing such as pawn structure hashing, where it >might be possible that two pawn structures could have the same hash key, and >this might go undetected. > >You should get the same variation (and same score) returned unless: > >1) You do any of the above. > >2) Your move ordering system has anything to do with positions seen previously, >for example history heuristic and possibly killer move heuristic. > >You should be able to take the "if (score >= beta) return score;" line out of >your program and have it return the same scores and same PV's, in signficantly >longer time, unless you do any of the above, or you have a bug. > >bruce I made this list seem exhaustive but it isn't. There are lots of ways you can introduce inconsistencies (which you may wish to leave in), but in a pure algorithmic sense alpha beta must return the same result as min-max. bruce
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.