Author: Michael Yee
Date: 06:18:29 02/21/05
Go up one level in this thread
On February 21, 2005 at 04:52:56, Uri Blass wrote: >On February 21, 2005 at 00:17:09, Michael Yee wrote: > >>On February 20, 2005 at 10:39:47, stuart taylor wrote: >> >>>On February 20, 2005 at 10:36:22, stuart taylor wrote: >>> >>>>http://web.telia.com/~u85924109/ssdf/list.htm (any better?) >>>> >>>>According to this most recent ssdf list, doesn't it look like the past 2 >>>>Shredders are the biggest challenges? >>>>Not only that, but often, Shredder 7.04 got better results even than Shredder 8, >>>>and here, it is only one point behind, which is nothing. >>>>So the question is, has Shredder 9 got out of that hole, and become more >>>>noticeably better, or not? >>>>And, not always have upgrades of chess programs been stronger. >>>>S.Taylor >> >>The CSS rating list shows Shredder getting stronger with each version: >> >>http://www.computerschach.com/rangliste/rangliste_ewig_ssdf.htm >> >>Different lists just show that computer chess has some randomness and it can >>take a while for accurate strength estimates. >> >>Also, I think there are some subtle biases in the elo system that depend on the >>strength difference between opponents. See, e.g., >> >>http://www.chessbase.com/newsdetail.asp?newsid=562 > >Thanks for that link. > >Some of these ideas are used by the israeli chess federation for years but I do >not like them. > >1)Using the linear formula: >Surprise is always possible so it is never logical to give 0 points to the >winner. > >You can claim that there is a better formula but we should try to find the best >formula and not linear formula. > >2)Using faster time control at the same rating. >It is done in Israel for some years and I think that it is not fair because >there are players who perform worse at faster time control. > >I played in the past in some tournaments at fast time control (30 minutes per >game) and from experience I always lost rating in them. > >Results is simply that I stopped playing at them and I guess that my rating at >30 minute/game is probably 100-200 elo smaller than my rating at long time >control. > >Uri With respect to (2), I guess the idea is that using more data (even if noisy) can improve your estimators. It allows you to use some general trends from the population (e.g., that good rapid play tends to correlate with good long play) to hopefully remove some error from an individual's estimate. In fact, you could also other marginally related data like "rating of your father" which could possibly improve predictive accuracy some very small amount. What I find interesting is that since you know your own true strengtht pretty well, you know for yourself that using the extra information nudged your rating in the wrong direction! Michael
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.