Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Dangers in CC - SSDF: Terminology, Statistics

Author: Maurizio De Leo

Date: 07:11:15 02/21/03

Go up one level in this thread


>>I had studied the topic before. And I have taken enough classes with maximum
>>mark in statistics to understand what we are talking about. Yes, errors are
>>always possible, but I seriously doubt to have made a "TERRIBLE" one.
>
>Let me explain. I believe that you have studied even more than myself, but you
>make a terrible basic mistake and that gives reasons to think. Only if you >could omit such basic mistakes you were a real expert. Therefore I call it >terrible although you might be a real expert soon. It's really funny that lay >often doubt how experts discover why someone can't be called expert. They say >but he knew all the necessary stuff and still he was called not-yet-expert.


But you don't say what is this "TERRIBLE" mistake. Can you please point this
mistake out, so that everyone (me included) can understand it ?


>>Ok, so for you "Shredder" is a false answer also to the question "who is SSDF
>>number one" ?
>
>Yes and no! You see, that is exactly what I told you. You think in black&white
>routines. But then you cannot understand what I am talking about. Of course
>Shredder is a possible first. But not a forced first.

"Yes and no" isn't an answer to me. Yes, I think in black and white, because so
does all the computational mathematics. I'm not interested in the philosopical
aspects of the SSDF list, I'm interested in the mathematical ones.
So I only will answer to precise and clearly explained logical flaws in my
argomentation, not to philosopical ramblings.


>>I don't agree. Let me make an example :
>>
>>1) Take the result of a Round-Robin Tournament, be it a human or computer one
>>2) Put them inside elostat
>>
>>You will have a list like the SSDF one, likely with high confidence intervals,
>>and likely it won't be possible to deduce from there who is the best.
>>
>>Neverthless, noone would question who was first in the tournament.
>
>
>I would have no objections if the SSDF would make a tournament for the 8 new
>entries. But NOTE: then you lose all 2700 hype. You begin at 0. That is the
>reason why they do a tournament  - as you mention - but they embedded t into
>their list. That alone is more religion than stats.

I don't care about the 2700 Hype. Being an expert you should know that the only
thing which counts in ELO numbers is ****"difference of ratings in the same
pool"****

What they did is this. A big big tournament with thousands of games, where
partecipants are ordered based on their performance ratings. The winner of this
tournament till now is Shreddder. The best player is ....... with .......
confidence.
At the place of the dot you can choose to place a lot of names with an high
number for confidence or the opposite (less names and less confidence).


>>Shredder performed slightly better than the other programs and it is at the
>>moment the number one.
>
>You fell into the trap of religion. You wat to have a umber one. But the actual
>data do NOT allow to tell.

The Data allow to tell who is number one, as the 16 games between Kasparov and
Kramnick allowed to tell who is the champion, even if they aren't proof for who
is the best.

You seem to miss a VERY easy point. All the times in all the sports we declare a
champion based on a FEW result, which can't PROOF that the champion is also the
best.

We can never proof who is the best, we can be sure 90%, 95% or even 99% but
never 100%.

So your suggestion to put togheter the first 5 programs isn't logical. Why
shouldn't we elevate the confidence range from 95% to 99% and put togheter the
first 12 programs ? Or why we don't elevate it to 99.99999999999% and get so big
intervals that we have to include almost all programs ?


>>It isn't for sure the best (although it may be),
>>but it is the number one.
>
>Hehe. Only if you could solve the squaring of a circle. You still dream of
>something the data simply can't give you. But you close your eyes and say yes >I can see the new number one. I call it a delusion.

You can call it as you like. But you should SHOW why it is a delusion or the
squaring of a circle, not just SAY it it so.


>>I would be really astonished if someone couldn't get this.
>>Think about the Fide list.
>
>Also this is already solved. FIDE has much more data and over the years. SSDF
>has only versios of one year length. And then finito. Then you begin at zero.

FIDE has about the same number of data for each player. I seriously doubt that
human players play more than 1000 rated games in their careers.
Moreover, SSDF has also computers from much more than 1 year ago, so your second
sentence is just plain false.

>>Who is first in a competition is something that we can always say for sure,
>>except when there are ex-aqueo for a not small enough measurement unit (which is
>>not the case in SSDF).
>>Who is best is something that we can assert with some confidence probability,
>>confidence probabilty that raises the more the gap between the first and the
>>second is.
>>
>>*************************************
>
>
>No, stop please, you begin to drift apart. There is a totally different method.
>For example SSDF gives different opponents!  IF then the progs have only 8
>points difference THEN prove me that NOT the different opponents did that!
>Ready?

Of course the different opponents may have been the reason for the 8 points
difference. This is why we have confidence interval and can't say for sure who
is the best. But the winner is still the winner.
In the world soccer cup every team has different opponents, so in every swiss
chess tournament. But there is a way to say who is the winner, although he may
not be the best.
It is really striking simple, how is possible not to get it ?


>>I was searching throughly for any personal attack in my letters and I really
>>have to say that I didn't found any. I was on the contrary pleased that you
>>seemed wanting a calm discussion, and I will hope that this desire is still firm
>>in you mind.
>
>Don't worry, brother, here is your insult. You made a statement that implied
>that usually I would NOT talk about topics but only about personal things. >Would you deny that? And that is an insult itself! Therefore my reply. Now >what is with your grandmother, I am very concerned.

Oh, I see. But that wasn't a personal attack. Simply by reading some of your
thread with Terry McCracken and another people whose name I don't rember I saw
that you (I mean all the 3) never talked about the topic of the subject, but
were always calling names : "kid", "ignorant" and so on.
I think the thread are still stored somewhere.
So that wasn't a personal attack. On the contrary I was very pleased that you
seemed wanting to talk about the topic and so I wrote it.


>>I never beated my grandmother which is unfortunately dead long ago. I don't >>find nice to cite other people relatives, but I'm not easy to be upset, so no
>>problems.
>
>
>What is with your second grandmother? Or did you have only one? That would be
>biologically highly interesting. <sigh> But I see that I am probably talking >to a very young person. Otherwise you knew what my question meant. <sigh>
>This was my last lesson for free. Thanks for your challenge.

My second grandmother was dead even before I was born, so I never knew her. I
still don't find nice to talk about other people relatives.

I didn't make any "challenge" to you, neither you gave me any "lesson", so I
don't know what you are talking about. Maybe your sub-optimal use of the english
language is related to why I didn't know what your question mean.
So you can as well reformulate the question in plain english, and maybe in the
meantime also try to reply to all the points that I made in my two letters and
that you avoided.

Maurizio





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.