Author: Alex Szabo
Date: 09:17:24 01/05/04
A self-play experiment between crafty version 19.6 (19.6) and crafty version 19.6 compiled with -DFUTILITY (19.6b) resulted in the latter being stronger than the former by 12 (plus or minus 2) elo rating points. The score was: 19.6 won 5,612 games, 8,109 games were drawn, and 19.6b won 6,279 games (for a total of 20,000 games in the experiment). A 2nd self-play experiment between crafty version 19.6 (19.6) and crafty version 19.6 compiled with -DDETECTDRAW (19.6c) resulted in the former being stronger than the latter by 21 (plus or minus 2) elo rating points. The score for the 2nd experiment was: 19.6 won 6,735 games, 7,734 games were drawn, and 19.6c won 5,529 games (for a total of 19,998 games). Note: The 19.6c version included the following source code changes to evaluate.c: (1) line #28 ...drawing=2 --> ...drawing=0 (2) line #53 ...!drawing... --> ...drawing... (3) line #3685, #3701, #3720, and #3735 ...max... --> ...Max... The experiments were run on a Dell Precision Workstation 450 with 2x2GHz Xeon (Pentium 4) processors running Fedora Core 1 (Linux). Xboard 4.2.6 was used to control the matches with the option "-tc 0:20 -inc 1" (time control set to "20+1" or 20 seconds per game plus 1 second per move). The 1st experiment was started 8th Dec and ended 15th Dec and took about 168.5 hours to complete; the 2nd experiment was started 15th Dec and ended 22nd Dec and took about 167.2 hours to complete. Additional details: (1) 4 matches were run at the same time (2) pondering was on (ie normally 8 crafty processes were competing for CPU time) (3) hyperthreading was disabled (4) use of 3 & 4 piece tablebases was enabled and cache set to 32M (5) learning was disabled (6) there was no opening book; games started from 10,000 different opening move sequences compiled from a database of grandmaster games (*) (7) each starting position was played twice with players reversed (8) all other crafty settings were default (9) the executables were compiled with "icc -O3 -march=pentium4 ..." (*) I started from gm2600.pgn (from ftp://ftp.cis.uab.edu/pub/hyatt/common) and truncated the games at the earliest point that the move sequences could all be unique (or 200 bytes max ascii move sequence length) and deleted any duplicates. I got 20,326 unique move sequences from which the 10,000 are a sample. It is possible that some of these starting positions are the same because of move transposition.
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.