Author: Bruce Moreland
Date: 21:53:39 06/15/01
Go up one level in this thread
On June 15, 2001 at 23:19:53, Robert Hyatt wrote: >On June 15, 2001 at 22:23:38, Mark Young wrote: > >>Nothing more really needs to be said, but you guys can debate more among >>yourselves. To restate no ratings system is absolute the rating numbers be it >>2500 or 3000 mean nothing in themselves. What is important is the difference >>between two ratings in the same ratings pool. Anyone knowing how ratings work >>would never bring the word absolute into the discussion as Bob has done. >> >>Ratings and whichever system is used to calculate the numbers are simply a >>measure of dominance in the ratings pool. If the system is working correctly it >>should be able to predict with a high degree of certainty the outcome of a >>series of games between two players in the same ratings pool. Even if the >>players have never played each other before. >> >>The problem with discussing this issue is some do not understand what ratings >>are, and how they work. As when Bob stated all players in the ratings pool must >>play each other a equal number of times for the rating system to be accurate. >>This is just incorrect. Thing start to become more ridiculous from there. > > > >It is not incorrect. Go pick up _any_ book on sampling theory and give it a >read. I gave an example of why your statement above is _wrong_. Take a group >of 20 players. players 1-10 play each other all the time. Players 11-20 play >each other all the time. Comparing ratings for the 1-10 group will give a very >good prediction of how they will do in a game vs each other. Same for the >other group. But then use those ratings to predict the outcome for player 1 >vs player 20. Won't work. Yes, the assumption is made that one opponent is as good as another. If you end up with isolated sub-pools, anything can happen. If a human figures out how to smack Crafty by playing it a zillion times, and ends up with a huge rating, it doesn't imply that they will score a certain percentage versus another human with a particular rating. I'm a good example of this myself. I can beat the ICC Bachs and Wimps, but I would do much less well against humans, which is one reason I don't play them. >If you have one player from the first group that plays one player from the >second group a lot, that tries to "couple" the two pools. But not very well. >However, if _everybody_ plays _everybody_ then you can predict the outcome >of any two players pretty well. Yes. If you have one large pool, you should be able to predict the results. But if people have worn ruts into the ground playing only certain other players, something weird might happen. The SSDF opponents are a small pool of players with common characteristics. The characteristics are different from the characteristics of humans to some degree. They play thousands of games against each other and only each other. It doesn't have to be true that you could throw a few accurately-rated humans into the SSDF pool and nothing would change. >That was what I said. That is simply fact. > >Which means you can not predict ratings for players that played before a formal >rating system was in place. Any more than you can compare a player of today >with a player from 40 years ago and use their ratings to predict how a game >between them would turn out. That's different. It's kind of like a cline. Actually, I don't know exactly what a cline is, but the idea is that if you have some birds that live in Maine, and some other birds that live in Florida, and you have the same kind of bird all up and down the coast, the birds in any given area may be able to breed with the birds in any adjacent area, but the birds in Florida might be enough different from the birds in Maine that they can't breed. You end up having one species up north, and another down south, but the line between the species is very blurry. You might be able to compare time periods by examining results attained by people that played over long periods of time, but people change and styles change and what it takes to play the game at the professional level changes. I'm sure that someone has researched this thoroughly though, so if I expressed a real opinion about I would probably be vulnerable to anyone with an education. bruce >I don't see why this is so hard to follow...
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.