Author: Uri Blass
Date: 05:49:54 07/17/02
Go up one level in this thread
On July 17, 2002 at 07:11:20, Sune Fischer wrote: >On July 17, 2002 at 06:27:20, Dann Corbit wrote: > >>On July 17, 2002 at 05:56:52, Nicolas GUIBERT wrote: >> >>> >>>>After reading the posts of others, I think that one thing should be made clear. >>>>I think you should decide what kind of tournament this is going to be, then >>>>develop the rest of it around it. For example, if it was going to be a "world >>>>championship", then (IMO) the goal is to determine the best engine in the world. >>>>If that is your goal, I would expect longer time controls... >>> >>>NO NO NO >>> >>>If you want to try and determine the best engine, do not play at longer time >>>controls !!!!!! >>> >>>BUT... Do play a lot more games with shorter time controls. >>> >>>Then you will get a more reliable result ! >>> >>>I have never understood why computer chess competitions are so slow. For human >>>players, this is understandable. For computers, it is just nonsense. I believe >>>that this has to do with history. The first competitions had to be played slowly >>>because the computers were terribly bad at Chess (low depths)... But that was 20 >>>years ago ! Computer science evolved quite a bit since then. >>> >>>Do you like playing random tournaments or do you really want to get an accurate >>>picture of the computer chess landscape ? >> >>Absolutely. The biggest problem (as I see it) is the quality of the games with >>the longer time controls. After all, we won't see any brilliancies like these >>at G/90 or slower on fast hardware: >> >>The true beauteous wonder that is "lightning chess" makes me pine for the days >>when all games will be run at G/1 minute or faster. > >It's not that big of a problem IMO, you get a lot of games with G/5, so even if >2 of 40 are flukes it won't matter, if you get 20 of 40 flukes then chances are >they spread evenly between the two contenders, so again it won't matter. >If you instead play two G/120 games, and one of them is a fluke, it matters a >lot. > >In other words: Say I give you two hours, you need to figure out if an >improvement you have made to your engine is really an improvement. >Now do you play 1 or 2 slow games against a third opponent, or do you prefer 20 >blitz games? >I would _always_ pick the blitz gauntlet, for better statistics (more games + >more opponents). > >I agree it is better (more interesting) to play 100 slow games than 100 fast >games, but that is never the choice we have to make :) > >-S. It is dependent on the change that you do. If you expect the change that you did to the program better only at long time control then you are going to play at long time control. 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.