Author: Thorsten Czub
Date: 14:05:25 03/06/01
Go up one level in this thread
classical problem: if they really run their basement tournament with ots of games and only ONE version of the program the programmer has a new version and the results generated by the guy ur not helpful anymore. nobody knows if the pattern the guy has outplayed for the old engine is also working for the new one. this is catch 22 or quantum-problem: in the moment you measure direction of a particle you change speed, and then you cannot know about speed. in the moment you measure speed you change direction. you canot know speed AND direction. same with programs. if you test too long, the programmer has a newer version and often cannot use your "results". therefore your test should be both: fast, precise and helpful ! when quantifying needs 100 times more time, and is more shallow, it is not as helpful as sample games. its very different depending on program and people. also: i know that many programmers let their engine X play against x+1 to get an idea about the strength. i also know that this can lead to contraproductive results, depending on the program. with some programs it gets better results. with others you destroy more than you increase strength. over all you have to make your own experience to find out. and often your prejudices change. personally i do not think playing x versus x+1 is a good hint. IMO it makes more sense to play x versus a,b,c,d... and let x+1 replay from the same openingposition x used against a,b,c,d,... and watch out the different behaviour.
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.