Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Chess strength of these programs?

Author: Thorsten Czub

Date: 07:11:01 03/15/98

Go up one level in this thread


>Sorry, but I think some common sense could help to keep things apart
>which should not be mixed up completely in that way.

>1. We have empiric facts.
>Here Enrique is completely right.

We is a group. I am not in this group. I am not a member of WE.
I don' the same results.

>a) the SSDF autoplayer results

I don't have this autoplayer nor do I know how the results were
generated.
As you have posted in another thread yourself,
one can easily manipulate ratings by emphazising WINNING lines in a big
book.
If you lower the weightings of LOSE games and INCREASE the weigthings of
WINNING games the book has still the same size but the learning
weightings let the program play DIFFERENT lines than a program WITHOUT
these weigthings.

How do you know that Fritz playing in sweden does not have special
learning files in his book, I mean: special weigthings generated from
autoplayer-games before ??
I knew this from crafty. Bob and Moritz told me once, you don't only
have to use the BIG BOOK, you also have to input the learning file.
I did and crafty chose different lines than before.
As far as I can see there are not many ways to proof that a book is
different when there are alternative lines.
The commercial book has the SAME lines but the ssdf-fritz could choose
only A FEW after lots of learning. How can we be sure that fritz5 in
sweden is not tuned this way as yourself have explained in another post
?!


>b) the games Moritz Berger played earlier which had *very* clear results
>and had *nothing* biased or questinable about them.

I cannot reproduce them. I cannot get fritz leading with 70 %, not with
60% nor even with 50% winning scores !
As long as I have different results , I believe in my own results asa
long as I see a mistake in my experiment.
You can call this biased. But I do better believe in my own data when
somebody tells me he has seen an UFO. I don't say he is lying. But I
have not seen this UFO.

>c) some game and test impressions of my own I told you and others *long*
>ago.

The same with your results. They are nice, but do not overrule my own
experiences.


>I then said I regarded Fritz5 as a Top5 program. And this from my
>P133/12Megabyte hash games and tests!!!

Pentium 133 ---- >>> Pentium MMX 200 is a difference.
12 ----->>> 44 MB hash is a difference too.

The increase is NOT linear. And any program reacts different.

Whatever you conclude from your observations, and nothing against your
experience, i cannot behave the same way.

Moritz e.g. always posted 70% or more in games Fritz5 vs. CSTal.
This is also ridiculous. I have a commercial CSTal and a commercial
Fritz (that was installed from the Junior CD) and fritz gets 44 MB hash
and his learning book and I don't get a better score fritz vs. CSTal
>=50 %.
I really don't know how you get your results. Anyway. We cannot come
into a peaceful data correlation.
You believe my results are exceptions. I do believe the same from your
data.

You knew all this very well and
>since quite a while.

Right.

>2. We have one more empiric fact :-)
>You *never* in all our private and public discussions, on phone or in
>the internet, would admit that Fritz5 might be a strong program.

This is not true. I would post it - if my results would reflect this.
But my results are horrible against fritz.

> The
>reason was easily obvoius to anyone talking with you: you don't like how
>it plays and you don't like the basic ideas behind its playing.

This liking or feelings have nothing to do with my 2 PC's. They don't
generate different games because I do like something or not. I am not
Uri Geller.
Your idea is not very constructive. You believe in patterns I don't
share.
The pattern is called: I would let overtake personal emotional feelings
over emperical data. I don'T do this. You believe I would do this. But I
don't.
I can differenciate between feelings and logic/ratio.
Also in live-tournament like aegon or paderborn or paris you can't
"selfbetray". If I would have done so, betraying myself with not
accepting data, why was I right in those live-tournaments ??!!
In my opinon you want to explain simething emperical with emotional
bias.
This is not ok.
You prejudice something you don't know.
I do not tolerate this behaviour.

>
>3. We have contradicting game results.

Here I agree with you 100 %. We have. I am in this WE-group !
But you make an emotional thing out of it where it has maybe algorithm
or technical background.
Unless we have not clearified this contradiction, I do not stop saying:
fritz plays very weak and also not > 50 %.
It IS not playing >50 % on my machine. And my machine plays very well.
it is not broken.


>
>Most of them all show in one direction:
>
>Moritz games

he has the autoplayer fritz !

>SSDF Games

They have the autoplayer fritz !


>Hallsworth games

As far as I know eric often overtakes undoubtful results / data from
other sources. I am not sure how he got the data in the fritz case.

>My engine games (on meanwhile a P200MMX with 64Meg)

Don't understand this sentence. In a few sentences later you said:
I have a pentium 133. Now it is a p200 MMX. This are 2 different worlds.



>ICC server games on faster time controls reported here.

We don't talk about faster time control. I talk about 40/120.
I am not interested in any rating on faster time control. I don't play
those games...



>
>These are only those I see at one glance.
>And they all show a similar playing strength at the very top.

They show that ONE parameter is different:
I do use the commercial NON autoplaying system and THEY use the
chessBase autoplayer. THIS is the main parameter that is different.
Is it therefore so far away to believe that THE FRITZ-AUTOPLAYER
is part of the contradiction ?!

>On the other hand you have been playing some games with other results.
>At least you can imagine Fritz5 to be 40 Elo stronger than Fritz4!

Maybe . But maybe not.
The problem is mainly we all recognized fritz a little better in
positional aspects, but we don't recognized a playing strength increase.
You and Moritz did.
I have not done this.
Peter has also not seen it.

You an Moritz work with/for ChessBase for a couple of years. So you are
pro Fritz.

Peter and I have different emotions. And emperical experiences.
Don't mix this up.
You cannot reduce the whole topic into an emotional problem.


>Now compare the data and do what you normally always did and did very
>well:
>get a realistic impression! I agree with you that if looking at games
>Fritz *would* always play like a dumb nut, it could hardly get a high
>rating. But whoever would say that Fritz did so would have no idea of
>chess! Fritz5 simply does not have to play chess the way you or i
>undertsand or like. It just is designed to calculate fast, play strong
>and win.


But it does not win here in Luenen. Maybe Schach Niggemann has not given
me a working fritz CD or a not working Junior CD. Whatever. I don't
know.
All other programs play very normal on my machines.

>4. Finally we have taste.
>Here we are quite close.
>Just like you I would never prefer Fritz5 to Hiarcs6 or Rebel9 based on
>some perfomance Elos. And even Junior's often cool and calm positionals
>moves attract me more often than Fritz5s. Nevertheless I can admit that
>it is strong, and nevertheless I can like the high aestetic value of
>many of its deep combinations or tricky denfending moves.

But this is not the problem i have. I don't have a taste evaluation. I
don't have the emperical data I would need to say: Fritz is a strong
program.


>So I ask you simply to open your mind and leave one of the few really
>fat prejudices you had since I know you.

The prejudices I have do not let my machines play different.
I am an idealist. But I normally believe a machine runs like it runs,
without beeing so easily manipulated by ideas.
Erdstrahlen ??


>As I told you in a private letter, I often learned from your well done
>obeservations concerning strength and playing style of *many* programs.
>I gladly repeat this in public. And that you were one of the first again
>telling us how strong Nimzo98 really is...

Nimzo produced from the very first day very good GAMES and results. My
taste and my data was against my prejudice. My prejudice said: I don't
like Nimzo.
My data said: it is strong. My taste liked the games.
So I instantly posted and emailed:

NIMZO IS STRONG.

I guess this shows in example that i do not let prejudice overrule my
data or let prejudice bias my data.

In fact this proofs that your idea that I am judging fritz with my
emotions and forget the data is not true.

My machines had no problem to generate normal nimzo game-results.
My machines have problems to generate fritz results. It seems:
my machines have emotions and do not like fritz.



>So give in for once and let's drink a cool, blonde Pils when we meet
>next :-)

Hm. I cannot give in since I don't have data that proofs your data.
Sorry.
We can - on the other hand, drink this pils.


>You know how this works, giving in? :-)))

I know it. I have done this before. I have given in in the NIMZO case.
In other cases too. I don't want to bore people with repeating these
many cases I have changed my opinion and I have told in public that I
was mistaken.

In the moment Fritz5 plays 40/120 vs. commercial tal.

1st game was draw.

I get completely opposite results Moritz gets.
HOW ?

>Yours Dirk



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.