Author: Robert Hyatt
Date: 07:10:08 11/25/03
Go up one level in this thread
On November 25, 2003 at 06:57:05, Sune Fischer wrote: >On November 24, 2003 at 23:19:27, Robert Hyatt wrote: >> >> >>Yeah, but when comparing humans, there is no decimel point in my >>comparisons. RE: Fide ratings. >> >>:) > >There are two questions here: > >A) how strong are the two engines in relation to eachother? >B) which of the two engines are stronger? > >You want to know A, you always want to know A, and therefore you care very much >about #draws. >Knowing A is much better than knowing B, by knowing A we also know B! > >Knowing B however is 'enough' and much easier to answer in general. There we don't agree. Perhaps in the case of computers, you play 1000 draws and then lose a single game. Later you discover that you lost because the hardware was screwing up on an ADD for a specific combination of 2's complement negative numbers. And produced a positive result (this happened to me on a Cray, once). But that loss makes you look worse. When the 1000 draws suggests equality. > >The only thing that keeps this from being really interesting in practise, IMO, >is that selfplay matches do not necessarily give a valid picture of strength >relations, which means answering B isn't enough to see if there is progress. >By playing against a third alien engine we are going to need an answer to A to >see if there is any progress. > I don't think that is good enough. The "alien engine" might not hit on a weakness in the new version that the old version does not have. IE if you play against an engine with weak king safety, your endgame bug might never cause a problem since you mop him up tactically due to the exposed king. >-S.
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.