Author: Andrew Williams
Date: 12:43:46 07/17/98
Go up one level in this thread
On July 16, 1998 at 11:15:03, frank phillips wrote: >Thanks to those who responded yet again to my request for help. I found the >major bug (I hope) causing the program to be so slow and thought I better ?own >up? just in case it might help some other newbie. My program now does 6 ply in >4 seconds and 7 ply in 30 seconds from the opening position on my K6-233 - >without much evaluation of course. Your comments helped me step back from the >detail I have been struggling with and identify the problem, which was a full >width search at ply[0]. To begin with, I developed the program without the >search to make sure that the move generator, makemove, unmakemove and in-check >detection worked. So to start with it effectively played a one move look-ahead >game by calling a function ComputerMove( ) and scoring the position at one move >ahead for all legal moves then choosing the best. When this was working I added >a Negamax and Quiescent search, but the function ComputerMove( ) was still >making each and every ply[0] move before calling Negamax. I shifted ply[0] down >to Negamax and the speed increased dramatically presumably becuase of the >alpha-beta cut-offs. The key was investigating how may nodes the program took >to reach each successive depth as Tom Kerrigan suggested. Now I can work on >some of the techniques (killer move, null move, hash) you suggested. What a >wonderful place this is, where a complete novice can get help and encouragement >from the experts. Two things: 1) Congratulations on getting this far! 2) The best step I took with my program was to interface it to xboard and thereby to the Internet Chess Club. This has two major benefits: (a) It is a *lot* of fun, as long as you can bear to see your program being crushed occasionally. (b) If you're like me and unable to beat your program, playing several games on ICC (or any ICS) gives you invaluable information about whether the latest "improvement" has actually made your program play any better. Obviously, the bugs that actually hang my program are easy to spot. But problems that reduce the (already lamentable) quality of its play are very hard to identify unless it plays against some competent players. Getting my program to work with xboard wasn't *too* hard (maybe 2-3 days plus debugging - others might have a better estimate) and the documentation with version 4.0.0 is now pretty clear. Anyway, good luck with your program. Andrew
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.