Author: Tord Romstad
Date: 12:24:44 07/29/99
Go up one level in this thread
On July 27, 1999 at 09:39:12, Robert Hyatt wrote: >On July 27, 1999 at 08:09:28, Chris Carson wrote: > >>On July 27, 1999 at 02:29:03, Gregor Overney wrote: >> >>>Using CM6000, I enjoy playing against different levels. What is the best way to >>>implement this "level control" feature into a Chess program? Truncation of >>>search depth? Adding randomness to score evaluation? >>> >>>If not too time consuming, it would be great to add such a "level control" to >>>Crafty? - Or, do you know about settings for Crafty that make it play like an >>>1800, 1900, 2000, and/or 2200 USCF rated player? >>> >>>Gregor >> >>Here are some things I have tried, let me know your feedback: >> >>Crafty (1200): Create Book = 10, Set Depth = 4 ply, Pondering = off, >> Learning = 000, no endgame tablebases >> >>Crafty (1400): Create Book = 20, Set Depth = 5 ply, Pondering = off, >> Learning = 000, no endgame tablebases >> >>Crafty (1600): Create Book = 30, Set Depth = 6 ply, Pondering = off, >> Learning = 000, no endgame tablebases >> >>Crafty (1800): Create Book = 40, Set Depth = 7 ply, Pondering = on, >> Learning = 000, no endgame tablebases >> >>Crafty (2000): Create Book = 50, Set Depth = 8 ply, Pondering = on, >> Learning = 111, 3 piece endgame tablebases >> >>Crafty (2200): Create Book = 60, Set Depth = 9 ply, Pondering = on, >> Learning = 111, 3&4 piece endgame table bases >> >>Variation on each rating can be made by using different opening books, >>setting king safety, setting piece values, ect... >> >>Best Regards, >>Chris Carson > > > >I don't believe those will work as expected. IE I don't believe that crafty >at 9 plies is going to play like a 2200 player. It will still find mates in >10+, and find tactical shots that a 2200 player won't find. > >I think the new 'eval' command (that lets you scale various parts of the eval >up and down) gives a more realistic 'feel'. And the 'extension' command lets >you tune down search extensions to stop the deeper combinations without making >it play like an idiot... I have a different idea about how to weaken a chess engine. Some day I will probably try out this idea myself, but in case anybody else is interested in trying out this I describe it here: My idea is to try to emulate human tactical oversights. This could be done by pruning _random_ moves within the search tree. For instance, you could choose some number n and immediately prune the next move your move generator returns every time your node count is a multiple of n. By varying the number n, you can adjust the strength of the program. This idea has a number of obvious refinements. To make the oversights more realistic, it might be a good idea to make the probability of pruning moves increase with the ply depth. It would probably also be interesting to increase the probability of pruning moves which appear to lose material (use SEE for this). Tord
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.