Author: blass uri
Date: 22:29:03 10/19/98
Go up one level in this thread
On October 19, 1998 at 23:03:08, Komputer Korner wrote: >On October 17, 1998 at 11:30:38, Jari Huikari wrote: > >> >>Could anyone of you find following statistics from a large amount of >>chess games? (E.g. from FICS or some Computer-Computer events...) >> >>-How often white/black checkmates the opponent? >>-How often either side resigns? >>-How often stalemates occurr? >>-How often the game is otherwise drawn? >>(mutual agreement, 3. repetitions, insufficient material, 50 moves rule) >> >> Thanks! >> Jari > > >Out of Knut Neven's Research Base of 1270187 games: >Checkmates= 6% >Resignations= 63.26% >Draws of all kinds= 30.74% >Stalemates= 0.05% >Other draws=30.69% > > >Some bases may have more draws, but in my base I deleted a lot of >repeat short draws that may actually have been between different >draws, but it wasn't more than 20,000 or so of those so that can't be >the reason. The actual reason is that GM vs GM games are less than >100,000 of the total so the so called GM draw is less prevalent among >lower rated players even though Knut's base quality of the opponents >is high enough to average 2445 of the players in the base that were >given a rating in the game header. This obviously is skewed because >almost all of the lower rated players in the base wouldn't have been >given a rating in the game header. So the bottom line is that it is >impossible to know exactly unless one was to manually look up the rating of >every player in the base and this is something no one will ever do. However >there is a way. We know that players at 1000 ELO against each other draw >less than 5% of the time and this draw % would increase until >at the 2800 level players draw nearly 80% of the time against each >other. So I guess some tests could be made to get the exact >equation but I don't have time to do this. I would guess that Knut's >base based on the draw % would probably average about 2300 ELO. From >looking at the numbers with some quick calculations which are by no >means accurate, I come up with a hasty formula where >% draws = (absolute value of (ELO/80))^1.11 note that this is changed from the >erroneous equation I presented on RGCC. I think that this formula does not hold for computer programs the level of best computer program is clearly better than 2300 ELO and there are less than 30% draws betweens computers. In shep's tournament there were 5 draws out of 21 and maybe there should be less draws because I am not sure that the game between fritz5 and Junior5 the was decided to be a draw is a draw(both programs did not have a plan and after 15-20 moves without progress a draw was decided not by the programs. I expect more draws when computers get better(because of hardware and better programs) but I do not see it. I remember that Genius3 did 4 draws out of 5 in the last world computer championship on a pentium90 and I think that there was wchess did the same number of draws. Now the maximal number of draws out of 6 games between top programs is only 3. Uri
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.