Author: José Carlos
Date: 00:45:09 12/10/99
Go up one level in this thread
On December 10, 1999 at 02:39:25, Bertil Eklund wrote: >On December 09, 1999 at 22:58:28, Robert Hyatt wrote: > >>On December 09, 1999 at 22:11:00, Len Eisner wrote: >> >>>The older programs on the SSDF list are underrated and the newer ones are >>>overrated. Why is that? >>> >>>Bob Hyatt's view is that the best of the current programs are about 2450. That >>>is at least 200 points less than their SSDF ratings. >>> >>>The older programs seem to be underrated by about as much as the new ones are >>>overrated. For example, the Fidelity Mach IV is only rated 2074 when it should >>>be over 2250. The old Novag Super Constellation is only 1731. I know it was at >>>least 200 points stronger than that. >>> >>>Len >> >> >>I don't think the mach iv was anywhere near 2250. It was at action chess >>(game/30/game/60) but not 40/2 >> >>What I think has happened is that newer programs blow older ones out, and >>artificially inflate the newer program ratings, and artificially deflate older >>program ratings... >> >>The older programs are not played against each other any longer, and the only >>way their ratings can go is down... > >Hi! > >There is no or very little inflation in the list. Everyone knows that todays >better players play a lot with computers and therefore can find the "wholes" in >there play (proved by you). I guess an experienced player earns 50-150 points vs >the one that haven´t played computers. One of the only later results is from >south-america where Hiarcs6, Rebel10 and CM6000 on P2-400 performed in average >2603 over 10 or 11 rounds. These matches with Rebel with very motivated >( prepared) players and double-increment time controls have nothing to do with >the normal way of achieving an established elo (in tournaments). I think Rebel´s >about 2500 under these circumstances is pretty fair. > >For the above most of the older programs haven´t moved much for years, but with >todays knowledge on chess-program they are over-rated to. > >Of course we should adjust the level of the list as soon as we have some proper >results. I think the list shows the difference between the programs in a rather >good way. Anyone can adjust the list to what he/she thinks is the correct level. > >Regards Bertil SSDF It seems to me that rating based on comp-comp cannot be compared to rating based on human-comp because of a matter of regularity (I hope this word means in english what I want to say in spanish :)), this is, a small playing strenght difference in comp-comp leads to a bigger ELO difference than the same playing strength difference in human-human. Why? Because comps are much more regular both in their strenghts and in their weaknesses. So, if you increase slightly the speed of a program, it would lead to a bigger ELO difference that what could be expected. So, the rating range between strongest-comp and weakest-comp will always be bigger that the rating range between human players of the same strength. The only reasonable possibility to get an accurate ELO for comps is a human-comp rating. Just my opinion. José C.
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.