Author: Peter Fendrich
Date: 14:01:25 04/25/03
Go up one level in this thread
On April 25, 2003 at 15:50:42, Albert Bertilsson wrote: >Hello! > >Of course I understand this, I meant the 2 to 1 and 3 to 2 numbers as >representation of the relationship. I never play less than 50 games and I >understand that it is not enough to make a good prediction. I have to find a >balance between testing often and testing good, and 50-60 games is all that I >can do over a night. The main point for testing is of course to find out if the >new version is stronger, the reason for having the rule of thumb was to have >some idéa to tell if the things I'm working on are worth the trouble or if I >should focus on something else. > >Still I thank you for pointing it out since there are many who take the result >from a 10 game match as a given fact, or even draw conclusions from fewer games. >I would actually like to see a greater focus on this since the normal person on >the people at best know about the latest man vs. machine tournament and thinks >that six games says it all. > >/Regards Albert Ok, I'm a strong believer of that you will get more reliable predictions from many fast games vs a variety of opponents than a few long games vs fewer opponents. With todays processors, close to 3Mhz and faster, bullet games will give this information. You can always verify it with longer games now and then. Of course engines relitive strength can differ depending on time settings but for most engines I think this is true. /Peter
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.