Author: Johan de Koning
Date: 23:54:01 08/19/03
Go up one level in this thread
On August 19, 2003 at 16:52:43, Sune Fischer wrote: >>> Maybe pridictability is even >>>a bad thing here, it disrupts the natural laws of statistics, making it possibly >>>even harder to test. >> >>I think now you're stretching it. :-) > >Actually, I was semi serious. :) I suspected you were. :-) >Suppose you do a gauntlet of 5000 games. >Now you change a small parameter and play again. > >So now you get 5000 more games, but of those 4800 might be exactly the same >because the change of parameter didn't affect them (you are determanistic so you >will get the same games under the same conditions a priori). >So in effect you only played 200 games. 200 games is not enough to see anything >so you have to play again, but now you get the same 5000 games as last time, >this is pointless! > >So how do you move on, how do you collect data without randomizing at some >point, be it in book or otherwise? Yes, you do need randomness for statistical analysis. But you need true random, rather then inherent variation or side-effects. If you play 1000 games program A versus B without opening books, you will get a lot of different games. Because (pun ahead :-) minute changes in timing accumulate until either program decides to go or not to go another ply. The rusult of this match does however not reflect the real world performance of A versus B, aka the population. Let alone A versus the rest of the alphabeth. ... Johan
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.