Author: Christophe Theron
Date: 12:20:10 10/25/01
Go up one level in this thread
On October 25, 2001 at 13:28:19, José Carlos wrote: >On October 25, 2001 at 11:30:04, Christophe Theron wrote: > >>On October 25, 2001 at 07:39:46, José Carlos wrote: >> >>>On October 25, 2001 at 06:57:29, Mike Hood wrote: >>> >>>>On October 24, 2001 at 22:03:49, Stephen A. Boak wrote: >>>> >>>>>On October 24, 2001 at 11:43:18, Joshua Lee wrote: >>>>> >>>>>>For Starters If Deep Fritz were that Magical 2700+ number Like the SSDF Claims >>>>>>Then Huebner wouldn't have Drawn Every Game of their 6 game Match >>>>>>Secondly With All Do Respect No Commercial Program Has Played As Many Humans As >>>>>>The Deep Thought/Blue Programs and Also The Number of Games Vs. Rating Average >>>>>>Is Unequal (Not as many games as Deep Thought) If you Suggest that programs are >>>>>>So Strong why Then Hasn't One of the Top Commercial's Put up so much Money as to >>>>>>Play Against a Top 10 Opponent and Not a Couple of Unknowns? >>>>>> >>>>>>Tiger Didn't Beat All GM's and I don't think they were very Strong GM's someone >>>>>>even mentioned that Tiger was Lost in One Position. That may not say Much but I >>>>>>would Consider Rebel's Achievement or Deep Junior's Much More Impressive. >>>>>>Rebel because of So many Games against Strong and well Known GM's Like Rhode and >>>>>>Scherbakov and Deep Junior for Beating GM Leko and Heubner , Drawing Everyone >>>>>>else Besides Kramnik and Lautier. >>>>>> >>>>>>8 Games are not really enough and 1 Tournament By no means makes a Computer a GM >>>>>>, They Can't Get The Title anyway, I would Like for this to be a possibility >>>>>>Then maybe someone would Try for their program to get it and we could Look to >>>>>>FIDE instead of SSDF . I hate that the list should be lowered by upto 200 >>>>>>points even by their own estimate the link is on their page. >>>>>> >>>>>>Another thing Tiger's Rating On an 866 Compared to the Speed Difference of the >>>>>>SSDF would Still Point to the SSDF's Given Rating for Tiger to be Wrong. >>>>>> >>>>>>Tiger is 2703 on a 1200 >>>>>>While 2788 against an average 2497FIDE On a Slower 866 Hmm Somebody is wrong >>>>>>Either all those players were lying about their rating or Could it be that the >>>>>>SSDF Is Off ... >>>>> >>>>>Curiousity leads me to pose some questions to thoughtful posters: >>>>> >>>>>Ever hear of natural variation? Do you think that a 2497 player plays at 2497 >>>>>strength (whatever that means) on each move, and across each game, no matter the >>>>>day or time or opponent or how well he is feeling? >>>>> >>>>>Ever hear of the uncertainty of measurement? What is the level of confidence >>>>>that a 2497 player is *actually* (whatever that means) a 2497 strength player? >>>>> >>>>>Can you accept random chance (natural variation) as a reason for occasional >>>>>exceptional results for programs or humans? >>>>> >>>>>Can you accept that measurements are all subject to some level of uncertainty, >>>>>some level of confidence less than certainty? >>>>> >>>>>If so, the above statement (prior poster) makes little sense. >>>>> >>>>>If not, I understand the dilemma and recommend a good introductory book on >>>>>statistics. >>>>> >>>>>Opinions are welcome, I have no problem with them. But do posters investigate >>>>>and try to learn about the subject they comment on, or are they curious to >>>>>discover what they may be missing in their view of things? >>>>> >>>>>Math is not a solution to everything. It is an often useful tool. It both has >>>>>its uses and its limitations. But to ignore it completely seems silly. Do >>>>>posters know they ignore some basic uses of math (often statistics) when they >>>>>post? Do they care? >>>>> >>>>>Just curious. >>>>> >>>>>--Steve >>>> >>>>Thanks, Steve. I often have thoughts like yours when I read posts with titles >>>>like "Beowulf is better than Deep Fritz on a 1.6 Ghz PC". >>>> >>>>What is the statistical background of the ELO rating system? >>> >>> As I've asked some times: is there a good mathematical way to measure >>>'strength'? What is 'strength' actually? Can anyone give a precise definition of >>>'strength'? Without such a precise definition we can't draw any conclusion at >>>all about players' strength. And if we want to draw mathematical conclusions, we >>>need a mathematical definition. >>> IMO, measuring ELO rating (which is defined by a mathematical formula) is very >>>different of measuring 'strength'. >> >> >> >>Strength is very clearly defined by the elo system. At least "relative strength >>inside a pool of given players". >> >>If you have a better definition (as you do not seem to be convinced by the Elo >>definition), feel free to submit yours... > >> Christophe > > I wish I had one, but I don't. As you said, Elo provides a good [enough] >definition of "relative strength inside a pool of given players". That shows >exactly the point I try to make: we can measure _realive_strength_ and >_inside_a_pool_. And even that is debatable, because we are saying >strength = results. > Ok, it's fine way to do it. > But it seems that some people talk about "absolute strength", and for that, we >don't have a definition, AFAIK. We could define absolute strength as the performance on a given set of positions, but that would be arbitrary anyway. The simplest definition is "the strongest wins more games". This defines strength in term of "offset" between two or N players. Christophe
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.