Author: Bruce Moreland
Date: 17:54:55 06/08/98
Go up one level in this thread
On June 08, 1998 at 13:16:34, Albert Silver wrote: >Well, I can't comment on what ICC was like in the past but I can say >that I have no understanding of the ratings in ICC. I have on more than >one occasion played opponents who were rated at least 200 points above >me and found myself rewarded with 1 (!) extra rating point after beating >them, then I'll beat an opponent rated 50 points less and get 10 points >or more for my efforts, so you can imagine what I think of the validity >of ratings there. The rating system itself is performing as designed. If each side has played at least twenty rated games, it will work appoximately as follows. If both players have the same rating, the winner gets 16 points, the loser loses 16, and no points change hands if the game is a draw. If one player is 200 points higher than the other one, the stronger one will get 8 for a win, and the weaker one will lose 8. If the weaker player wins, they'll get 24 and the stronger player will lose 24. If the game is a draw, the stronger player will get 8 points and the weaker one will lose 8. You gain or lose more or less points if the rating differential is more or less than 200 points. You'll never lose or gain more than 32 points for a loss or a win, and you'll never gain or lose more than 16 for a draw. If one or both players hasn't played 20 games yet, everything changes. The rating of a player who has yet to play 20 games can change wildly. If your rating is established, and your opponent's is not, you'll gain or lose fewer points than you'd expect. So my guess is that in your first example, you played against someone who was provisional, so your rating change as slight, and in your second example, everything worked right. Sorry this is so long-winded. >>I used to use a P6/200 on ICC, it peaked out at 2800 or so, back when >>this was the best anyone else was doing. Now when I run the same >>machine, 2800 is a disappointment, 2900 is more like it, and sometimes >>it exceeds 3000. >> >>The software is better now but not that much better. > >There is a funny possibilty now that I think of it: it is well known >that some players are really very successful against computer opponents. >Even when their colleagues who are rated 200-300 points more than they >do not do nearly so well. This would really mess with the rating system >as let's say a master scores a solid 50% against a 2800 rated computer >account and gains a generous amount of rating points which he promptly >loses to GM X who is only rated 2600 but who outplays him regularly. >This goes beyond the usual nemesis element in which a player due to >stylistic incompatibilities just can't seem to do well against a certain >opponent. In tournaments this happens undoubtedly but on a much smaller >scale (Player A will only encounter Player B about 3-4 times in a year >if that many), but on ICC where tens and hundreds of games are played, >this divergence is multiplied enormously. Perhaps therein lies part of >the explanation. I think this is exactly true, other than that the idea of any commonly occurring human player beating a 2800 computer 50% of the time in blitz is pretty optimistic. You do find cases where people will give away tons of points in super-fast reckless games against random opponents, then will play very slow and considered games against another opponent, and get them all back. I think it'd be really hard to do anything to fix this, short of attaching a breathalizer to everyone's computer. bruce
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.