Author: Robert Hyatt
Date: 11:32:52 10/28/99
Go up one level in this thread
On October 28, 1999 at 08:41:50, blass uri wrote: >On October 27, 1999 at 11:50:40, Robert Hyatt wrote: > >>On October 27, 1999 at 04:02:58, James B. Shearer wrote: >> >>>On October 27, 1999 at 01:30:58, Dann Corbit wrote: >>> >>>>On October 27, 1999 at 00:42:09, James B. Shearer wrote: >>>>[snip] >>>>> This is not accurate. The ELO system is based on a model of the real >>>>>world. Like all models this model is just an approximation of the real world. >>>>>This means that some potential opponents will appear stronger when playing you >>>>>than their ratings would indicate while others will appear weaker to you than >>>>>their ratings would indicate. By selectively playing those opponents which do >>>>>not play up to their rating against you, you can definitely raise your rating >>>>>above what it would be playing all comers. >>>> >>>>How is this selection made? Without hundreds of games there is no way of >>>>knowing who they are. >>> >>> Well for example you may play relatively better against computers than >>>other humans play against computers. Then if you play computers exclusively >>>this will raise your rating above what it would be otherwise. >>> >>>>If these persons/computers play others the ELO will balance out over time. >>>>That's how it works. >>> >>> The opponents you victimize can restore their rating to its proper >>>value by playing others, however your rating will remain elevated. >>> James B. Shearer >> >> >>That is true if (a) you are very selective in who you play and (b) you don't >>play many games to let them discover 'holes' in your book or evaluation. I have >>said before that the 'automatic' programs have a tougher time on ICC, _not_ an >>easier time. Because if they cook my book, they will cook my rating. A manual >>program will just simply lose and disconnect. And lose one game rather than >>20. > >There is another side of the story. >1)Automatic program can play fast time control against humans and get rating >from it when a manual program cannot do it. log on to ICC and look at all of crafty's games with GM players. 99% of them are 5 3, which is why I set that in my formula... that seems to be the preferred time control for most gm players, although a few still do 5 2. So 3 0 is only done on rare occasions and isn't a factor in long-term ratings. Manual programs can play 5 3 easily. I was manual for a long time and I was playing 5 3 and 5 0 with no real problems. Bountyhunter even plays _bullet_ manually and is very successful both vs humans and computers. >I believe that ban could play humans at 3 0 time control and earn rating >if it was an automatic program. >The fact that ban plays 5 5 games and not 3 0 games against humans is not >productive for rating. pretty much irrelevant since 99% of the games I play vs GM players are 5 3 or 5 5. Most GMs know to _not_ try 3 0 as it is nearly hopeless. > >2)The assumption that a manual program is going to disconnect after a loss does >not have to be correct. No... But the assumption that a manual program will play cptnbluebear, and discover that he _specifically_ plays for either long-term kingside attacks or else directly for draws, and after spotting this they simply won't accept further match requests. _Many_ manual programs have had this in their finger notes: "Note, if you are one of the players that seem to be almost able to draw at will, rather than trying to play chess and win, you will end up on my noplay list." >I understood that amir ban has no rule not to play again after a loss. I don't know. For the past month the _only_ opponent I saw him play was Crafty. No GMs no other programs... just Crafty. >I remember from a post of ban some monthes ago that in one case the opponent was >the one who disconnected after a win. > >Uri That is perfectly normal for me too. I notice that in a GM match the history alwasy starts off with a bunch of wins/draws, but generally when the GM finally wins a game, that is the last game of the 'match' and away he goes. Which is fine by me.
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.