Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Comments of latest SSDF list - Nine basic questions

Author: Bertil Eklund

Date: 15:20:00 05/31/02

Go up one level in this thread


On May 31, 2002 at 18:04:21, Rolf Tueschen wrote:

>On May 31, 2002 at 11:26:46, Uri Blass wrote:
>
>>On May 31, 2002 at 10:19:21, Rolf Tueschen wrote:
>>
>>>On May 30, 2002 at 22:20:02, Uri Blass wrote:
>>>
>>>>On May 30, 2002 at 21:55:21, Rolf Tueschen wrote:
>>>>
>>>>># Here follows a list of simple questions.
>>>>>
>>>>># 1) SSDF equips each time the new programs with the fastest hardware. Do we
>>>>>find out this way if the new engine is stronger than the older? No! Quite simply
>>>>>because the old engines could be as strong or stronger on new hardware.
>>>>
>>>>There are some programs that use more than one hardware in the ssdf list so we
>>>>can know the expected increase in rating from the new hardware.
>>>>
>>>>Things may be different for different programs but if the fidderence is big then
>>>>we are going to know if the new programs are better than the old programs.
>>>
>>>Mmhhm. Frankly I do not understand. Are you saying that we, the readers of the
>>>SSDF list, should be inspired enough to make our own conclusions, or do you
>>>argue that SSDF is doing a wonderful job?
>>
>>Both of them.
>>
>>We can make conclusions that the weaker programs are not good enough to be the
>>first because if they compete for the first place the ssdf is going to test them
>>also on the new hardware.
>
>?? Where did you get that from? It's not true because SSDF once tested the new
>Wch on older hardware.
>
>
>>
>>We can also guess the rating of the old programs on the new hardware based on
>>the rating of the old hardware and when the rating on the old hardware is lower
>>relative to the opponents then it will probably not be very high also on the new
>>hardware(programs may earn more from time but practical experience show that the
>>difference is small and not more than 10-20 elo for doubling the speed.
>
>Ok, you know what I wrote about "Elo" in SSDF. But how about the memory? Not
>much effect too? And also I think that we are not discussing SSDF IMO.
>
>
>
>>
>>
>>
>>
>>> And BTW what is your concept of "big"?
>>>What is big when you have a margin of +/- 30 points uncertainty?
>>>Could you please elaborate on that point?
>
>??
>
>
>>>
>>>
>>>>
>>>><snipped>
>>>>># 6) SSDF often matches newest progs vs ancient progs. Why? Because the
>>>>>variability of the choice of the opponent is important for the calculation of
>>>>>Elo numbers? Hence Kasparov is playing against a master player of about Elo
>>>>>2350? Of course not! Such nonsense is not part of human chess [as necessity of
>>>>>Elo numbers!]! Or is it that the lacking validity of the computer should be
>>>>>replaced by the play against weakest and helpless opponents? We don't know.
>>>>
>>>>
>>>>The ssdf does not play matches between programs that the difference in rating is
>>>>more than 400 elo and kasparov plays against players that are clearly weaker
>>>>than him.
>>>>
>>>>Uri
>>>
>>>Also this about Kasparov for me is rather a circle in the argument since
>>>Kasparov has the highest Elo number. By force he's always playing weaker
>>>players. Take Huebner, he has almost 200 (!) points less. But nobody would
>>>suppose that Huebner is absolutely outdated and a weak player. Although for sure
>>>he'll lose against Kasparov. But on the other hand would you support the
>>>practice of SSDF, I repeat one of the snipped points, where a program is weaker
>>>in all four categories and still is matched against newcomers with newest
>>>hardware? As we all know that in difference to human chess computers have a much
>>>more deterministic play? But again, we do _not_ know what SSDF is doing in
>>>detail.
>>>
>>>And finally how could we know anything at all if we have no validity? You didn't
>>>comment on this point at the beginning of my questions. Did you agree? But then
>>>I can't understand your confidence in concepts like for example "big" or also
>>>"400 points". How do we know that they don't test machines with differences of
>>>"400" points? If we don't know what SSDF numbers mean _at all_?
>>>And then the +/+ 30-50 "points" "uncertainty"... 2x50= already 100.
>>>Should I search for examples in the history of SSDF? I am certain that I could
>>>find such events of 300! Then my assumption about Kasparov would still be true
>>>because he ain't playing 2490 players.
>>
>>He is not but other humans play games with players that are 300 elo weaker or
>>300 elo higher then them.
>
>Yes, but only in OPEN! Not in tournament play.
>
>
>>
>>The ssdf almost did not play uninteresting matches(matches with result of 20-0
>>or 19.5-.5) so I see no problem.
>
>Uri, please excuse me if I become a bit more direct in my statements. You are
>completely misleaden if you say such a thing. BTW you are usually the one who
>analysed the game scores. What are you talking about when you say 20-0 is
>uninteresting? You must know like me that we can't compare computers strength
>with human chess player strength. Speaking in terms of stats:
>
>°° With the many thousands of _active_ human chess players you get well
>exactitudes of 10 or 5 or even less points difference between ranks, and _still_
>you have totally different individuals. However in computerchess with the actual
>testing of perhaps 10 programs you are either testing the same individual
>program or you are presenting artefacts, when you find 8 points difference and
>error margins of +/- >30. In other words you don't have a result at all.
>
>°° Normally the edition of the list in May 2002 had to be postponed or the
>ranking had to be completely designed with new ideas. The top of the May 2002
>list simply makes no sense. And the rest doesn't matter for the moment. The
>weaker programs are nothing but technical supplier for more points.
>
>Rolf Tueschen
>
>>

You proved it again! You didn't understand a single sentence! Impressing!!

Bertil



This page took 0.03 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.