Author: John Merlino
Date: 19:42:23 02/09/01
Go up one level in this thread
On February 09, 2001 at 22:21:43, Joe mccarron wrote: >Whenever I hear about a program trying to handicap itself to approximate a >specific elo, I always wonder how the programers take into account the different >cpu speeds. Does the computer do some sort of benchmark test when it is >installed and modify the personalities? Does the speed of the cpu not really >make much of a difference because of the design of the handicap? or shoudl I >just assume that when they say this setting plays about 1700 that means more >like 1800 on a new gigahertz+ and more like 1600 on a 486. >I also wonder how these handicaps might be affected by longer time controls. >That is I think most humans play considerably better at long time controls than >at short. If the computer had a handicap design that acted regardless of the >cpu speed wouldn't this mean that the computer would play considerably worse >than expected at the slower time control versus humans? >Of course I realize humans also might tend to vary greatly at different time >controls. Some players I know are good at short time controls but their game >doesn't seem to get much better with more time and vice versa. I was just >wondering how programmers approach these issues. I realize most programs done't >bother to try to approximate an elo but I'm wondering what some of the thoughts >on this are. Thanks in advance for any responses. Chessmaster takes four factors into account: 1) An estimate of the base USCF rating of the engine on default settings on the minimum spec machine (Pentium II-233). 2) A determination of the processor speed at runtime (done every time the program starts up, which is why you can get different ratings for the same personalities). This equates into a "bonus ELO" which is applied to varying degrees to all ratings. This variation is based on.... 3) How close the personality's strength is to 100. The farther away it is from 100, the less of the bonus it gets. 4) Additionally, personalities that have a fixed search depth get NO CPU speed bonus. This would seem incorrect, because some personalities could have a fixed depth of 20, and it would appear that a max depth of 20 would take a long time to reach. However, the only personalities that DO have a fixed search depth less than the maximum all have a depth of 6 or less, which is reached very quickly. Therefore, these personalities get no bonus. Of course, it's all just an estimation. In my opinion, the formula of (each doubling of CPU speed = 70 ELO points) is probably not as valid as it was 3 or 4 years ago when it was first introduced to the development team. Maybe it's more like 40-50 now, if that. But, the team has seen no reason to change it, yet.... jm
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.