Author: Robert Hyatt
Date: 09:42:19 07/13/03
Go up one level in this thread
On July 13, 2003 at 06:19:30, Sune Fischer wrote: >On July 12, 2003 at 23:58:24, Robert Hyatt wrote: >>>> >>>>If only high-rated players play high-rated players, you have _two_ pools and >>>>the rating difference between the two pools is meaningless. >>> >>>Not exactly. >>>even without swiss tournaments >>>2800 player can play against 2650 player and the 2650 player can play in another >>>tournament against 2500 player when the 2500 player can play in another >>>tournament against 2350 player... >> >>OK. That was my original point. The "new player in town" is not going to be >>2600 to start. He works his way up. > >It shouldn't matter where he starts as long as he doesn't get a rating until >"enough" games has been played. > >But since these testers has a pretty good idea that new program x version z is >better than the old version, it makes good sense to start out against the top. > >The only "problem" I might fear here, is the program x version z was built to >beat program y version w, then latter comes along program y version q, built to >beat prog x version z. > >This kind of inbreed might lead to artifically inflated ratings, but I don't >know much an issue this is, they seem to do really well against much lower rated >programs also. > >>>>>If shredder7.04(A1200) starts by playing 20 games against palm tiger14.9 and >>>>>20 games against Fritz3(p90) then I doubt if it is going to make it's rating >>>>>smaller. >>>>> >>>>>It has good chance to get 100% or almost 100% score in these games. >>>> >>>>But one draw will cost it rating points that it won't make up with 19 wins. >>>>The other point is that as it beats palm tiger, palm tiger is going down in >>>>rating also, which will mean that others that beat it will see a smaller >>>>rating improvement. >>> >>>If palm tiger's rating is going down then Shredder7.04's rating is going up. >>>I agree that in this case other that beats palm tiger may get smaller rating so >>>the influence is for both sides but I do not see a reason to assume that the >>>influence for going up is higher. >> >>Simple. Just see what happens on 19 losses and one draw. If the rating >>difference is significant, the one draw kills the other player's rating >>compared to the boost from the 19 wins... > >Sure, but how does this indicate inflation? >There is less margin for error if you have a really high expected score, >natually, but if you've earned that high rating through enough games, it >shouldn't be a problem. > >-S. I (and many others) believe that the Elo system works well for players that are pretty close in rating. It seems to work less well (in the case of computers) for players that are significantly separated in ratings. Starting at the top eliminates the bottom of the pool from the "war". If the bottom can't drop, then neither can the players at the top. So you get a new top. If you start the new (and strong) program at the bottom, he will drop _everybody_ as he goes up, and it would seem that this would result in the new player going to the right "differential" spot but that it might be lower than it would have been with a high start.
This page took 0.01 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.