Author: John R. Menke, Sr.
Date: 17:22:28 06/20/99
Go up one level in this thread
Hi Terry, Your match is very interesting, but I believe at least 30 games is required for reasonable statistical significance in determining playing strength (i.e. Elo rating), preferably more. Also, with selected openings the results might be biased relative to the usual method of play -- a random choice of openings by the two players. I used to conduct matches between different chess playing programs on a Commodore 64 (over 10 years ago), and concluded that what I could personally do would be insignificant. It takes a major effort to get statistically valid results. That's not intended to discourage you, but a friendly suggestion. To get significant results in a short period of time probably requires a lot of different people and computers running simultaneous matches. For example if you could convince a dozen different people to each run a dozen games with a particular set of CM6000 personality settings, a 144 game sample, you'd have something that the rating statisticians couldn't ignore so easily. Another approach is to look for some good "benchmark" positions and see how long it takes to solve them with different personality settings, or which personality gives the best solution within a limited period of time. I'm not a statistician, so my comments might be invalid!? It's fortunately all an enjoyable game and hobby for me, so hopefully no harm done in any case. ---JRM
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.