Author: Marc Lacrosse
Date: 02:27:24 06/17/05
Go up one level in this thread
On June 16, 2005 at 18:05:26, Vincent Diepeveen wrote: >You can distinguish yourself really a lot from the 1000 others, by playing some >serious time control. Like 40 in 2 + 20 in 1 + 15 (the fide time control that >all competitions and big tournaments use here). Of course WITH permanent brain >turned on at a dual core. 256MB hash or 400MB hash. Never cross the 512MB limit >in windows. This is asking for trouble. > >Nearly everyone is either playing 5 0, or 15 0 , or 60 15 or 40 in 15 or 40 in >25 or a very few do 90 0. > >Some real slow tournament level, yes even 90 30, will be very interesting to >see. I can of course from head already write down which of the engines are the >good blitzers among the engines you quote here. Everyone is doing blitz or rapid >time controls. Some serious time control, at least 90 30, and preferably 40 in >2, is interesting for me to watch your results. > Dear Vincent, Regarding the subject of timing, I completely agree with you that we miss longer time control games and tournaments. Being an active correspondence player myself I am pretty concerned with this, and with the wish to get as good games as possible. However there are not so stupid considerations you must deal with when planning a single-man operated tournament. Some of these considerations are bound to the duration of the tournament. Due to rapid evolution of both hardware and engines, it makes no sense IMHO to plan something for a duration longer than a few months. Imagine I organize a 20 engines tournament. There will be 20 * 19 / 2 = 90 individual encounters. As I cannot hope to have more than two encounters played per day, this will be going for 1.5 or two months which is optimal in my view. And I fear that the actual number of encounters played per day will be closer to one per day which is three months duration for the tournament. As I said in my original post, due to my specific situation, each individual encouter has to be played in less than 12 hours. So I have some choices I must face. Within this 12 hours lapse what should be played : two 6-hours games, four 3-hours games, or at the extreme opposite end sixty 5-minutes blitzes. If I choose the slower time control, almost identical to the traditional FIDE timing that you praise, each engine will have played 38 games at the end of the tournament, which is by far too few to have any statistical signification regarding relative strength of the engines. At the opposite, if each encounter is made of sixty blitz games, each engine will have played more than 1000 games at the end with a completely different degree of statistical signification. There is today a very interesting thread initiated by Volker Pittlik on the WBforum (http://wbforum.volker-pittlik.name/viewtopic.php?t=2849). According to its tests there is a very constant relative rating of different engines independently of the thinking time allowed in finding solution to the more or less positional arasan test suite. And in the discussion Volker said : " If I can choose between rating list based upon 100 games at blitz (if not run at a stupid speed of all moves in one minute or so) and 10 games at 40/2 hours I would choose the first. A completly different question is the quality of the games..." So we come back to the necessary trade-off between a few longer better games and a more significant number of weaker games. An additional point must also be considered. For personal reasons i wil have the tournament played under winboard + polyglot with polyglot delivering the same opening book to all engines. Although I take a book made of very good quality games, there is always the risk that one or another among the positions from which the engines will become to play is not equal, favoring one of the opponents. This arbitrary advantage will surely be less prominent if the number of games is higher because ti will be more randomly distributed. Another point : I am convinced that the quality of the evaluation delivered by a given engine is not linearly related to the amount of thinking time. It's more of an asymptotic curve. So if we compare the quality delivered at 3 minutes per move (the traditional timing 40 moves in 2 hours) to that we can get at some shorter time like 45 seconds per move, I think the quality of this latter will be closer to that of conventional play than to that of 10 sec per move blitzes. So I think a good compromise could be to change my planned timing to 15 min initial time + 30 seconds increment. In this case the maximal mean thinking time per move would be 53 seconds for a 40 moves game, 45 seconds for a 60 moves game and 38 seconds for a 120 moves game. This would result in a maximal game duration of 70, 90 and 150 minutes for respectively 40, 60 and 120 moves played so that I could hope to have 8 games played per elementary engine-engine 12-hours match. In this case, each engine will have played 152 games at the end of the 20-engines tournament, which is already an interesting global amount of games. Your opinion ? Marc
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.