Author: blass uri
Date: 22:42:28 05/23/00
Go up one level in this thread
On May 23, 2000 at 21:23:01, Dann Corbit wrote: >On May 23, 2000 at 13:47:51, blass uri wrote: >>On May 23, 2000 at 12:20:55, Rafael Andrist wrote: >>>On May 23, 2000 at 09:05:15, Robert Hyatt wrote: >>>> >>>>First, I don't think "90" is a "magic number". Bruce found that Ferret >>>>searching very shallowly could hold a rating > 2200 on ICC. At the time, >>>>his program was rated over 2800 in a real rating there. Which means each >>>>ply was worth much less than 90. So "N" points per ply, linearly? I think >>>>it is definitely possible. And yes, it is possible that some programs begin >>>>to 'flatten out' while others don't. I don't see anything that suggests that >>>>can't happen either. >>> >>>I think also that it's possible that the ELO is linear to the ply. But is the >>>ELO linear to the playing strenght? I mean that it would be very hard for a 1400 >>>ELO player to beat a 1500 ELO player but not for one with 2500 ELO to beat one >>>with 2600 ELO. >> >>This is not the idea behind the elo. >> >>It is easy to prevent it by not giving rating to people unless they play against >>players with known rating. >> >>Unfortunately players begin with 1350 elo rating(this is the case in Israel) so >>the rating is not correct and many strange thing can happen. >> >>It can happen that 1400 in one place is better than 1500 in another place. >> >>I suggested in the past to have a computer programs in every tournament when the >>humans have no rating and to use the program's result to decide about the rating >>of the players(it is possible to use a program that sometimes generate random >>moves so the beginners will have chances but not too often so the program will >>not lose all the games) >> >>I did not get a resonse about this idea and it seems that it is not important >>for the Israeli chess federation to have a right rating in the low levels. > >That's actually quite an interesting idea, since the computers should be >extremely consistent. However, in order to get an acceptable handle on ELO, the >computers would have to play an unacceptable number of games against the >participants if their ELO is actually unknown. > >On the other hand, if this procedure were used, eventually, you would get very >accurate figures that would level out even without the players ever having met. > >Perhaps they would agree to older programs on P90 hardware. Maybe something >with a rock-solid and extremely well known ELO of 2300 or so. Something with elo 2300 will not help much because it is going to win all the games against humans who start to play in tournament(there is a small probability that it is not going to win all the games but the practical case will be winning all the games against unrated opponents who play their first tournament). If the computer will have 9 wins out of 9 games we cannot use the result to give rating for humans. We need something that is weaker than it that will not get 100% or 0% and something that does sometimes random moves(to prevent humans to win the program in the same way). Uri
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.