How strong on a Pentium Pro 200?
James Reames
I am now the proud owner of a Pentium Pro 200 with 64mb RAM. I have two
questions:
1. What is the strongest WINDOWS based chess program?
2. How strong would it be on my Pentium Pro 200?
Thanks in advance for your answers.
Regards,
James Reames
ches...@mcs.com
http://www.mcs.net/~chessman
eric fitch
Chess Genius is a Windows Program and certainly the strongest. On a
PP/200? Well on the latest SSDF list, Genius 5.0 is at a 2420. This is
using the current standard P/90. I'm no expert bu I'd guess around 50
points more on a PP/200!???
Lawrence S. Tamarkin
- On 21 February, James Reames wrote;
- I am now the proud owner of a Pentium Pro 200 with 64mb RAM. I have two questions:
- 1. What is the strongest WINDOWS based chess program?
- 2. How strong would it be on my Pentium Pro 200?
- Thanks in advance for your answers.
- Regards, James Reames ches...@mcs.com http://www.mcs.net/~chessman
- It seem to me that Genius 5 is likely to rank the hightest based on the Swedish Rating List:
- (All on Pentium 90mzh)
- 1. Rebel 8 2462
- 2. MChess 6 2435
- 3. Hiarcs 5 2427
- 4. < Genius5> 2420
- 5. Rebel6 2412
- 6. <Genius4> 2411
- As can be seen from above Chess Genius is the only Windows program to place in the top 6. The next strong full featured program I am aware is Fritz4 which ranks #22 on the Swedon list with 2336.
- A very important creterior for selecting a chess playing program in my opinion however is it ability to integrate as a database. On this both Genius5 (4) and Fritz4 are very powerfull. But because Fritz4 is from the same company that produces Chess Base it is the best in that area. On your 200mzh Pentium pro, Fritz4 might be a lot closer to Genius or so close in rating that it should not matter to you unless you crave the chess strength on a purly status level. - But if that is true Rebel 8 would be the program to get. Even though it is dos based it has consistantly been the program to beat, and is marketed by a company that has recently put out some terrific databases for the money. Also a windows version of Rebel is quite possible as Rebel creator Ed Schroder's internet page [ http://www.xs4all.nl/~rebchess/quest.htm ] elicits user opinions for the next version of Rebel.
- Larry Tamarkin
- Chess database/program addict
Komputer Korner
James Reames wrote:
>
> I am now the proud owner of a Pentium Pro 200 with 64mb RAM. I have two
> questions:
>
> 1. What is the strongest WINDOWS based chess program?
>
> 2. How strong would it be on my Pentium Pro 200?
>
> Thanks in advance for your answers.
>
> Regards,
> James Reames
> ches...@mcs.com
> http://www.mcs.net/~chessman
Wait for Hiarcs 6 and boot your machine up in DOS or if you must
play in a DOS box but it will play weaker. Only tests will show how
much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
game match. We just need a sponsor to prove this.
--
Komputer Korner
The inkompetent komputer.
eric fitch
eric fitch wrote:
>
> James Reames wrote:
> >
> > I am now the proud owner of a Pentium Pro 200 with 64mb RAM. I have two
> > questions:
> >
> > 1. What is the strongest WINDOWS based chess program?
> >
> > 2. How strong would it be on my Pentium Pro 200?
> >
> > Thanks in advance for your answers.
> >
> > Regards,
> > James Reames
> > ches...@mcs.com
> > http://www.mcs.net/~chessman
>
> PP/200? Well on the latest SSDF list, Genius 5.0 is at a 2420. This is
> using the current standard P/90. I'm no expert bu I'd guess around 50
> points more on a PP/200!???
A correction to my response. The current rating for Genius 5 of 2420 is
for the DOS version only. No official results for the Windows version
from the SSDF. Still, it is certainly the strongest of the Windows
bunch.
Moritz Berger
On Mon, 24 Feb 1997 03:03:44 -0500, Komputer Korner <kor...@netcom.ca>
wrote:
<snip>
>Wait for Hiarcs 6 and boot your machine up in DOS or if you must
>play in a DOS box but it will play weaker. Only tests will show how
>much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
>game match. We just need a sponsor to prove this.
I will set up a match of a 2280 FIDE ELO player against Hiarcs 6 as
soon as it will be released. 6 games should be ok. Time controls maybe
30/60, machine will be a P5/166 unless I can get hold of a Klamath
machine soon enough ...
Moritz
-------------
Moritz...@msn.com
brucemo
Komputer Korner wrote:
> Wait for Hiarcs 6 and boot your machine up in DOS or if you must
> play in a DOS box but it will play weaker. Only tests will show how
> much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
> game match. We just need a sponsor to prove this.
Martin!
bruce
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
<33114B...@netcom.ca>...
> James Reames wrote:
> >
> > I am now the proud owner of a Pentium Pro 200 with 64mb RAM. I have
two
> > questions:
> >
> > 1. What is the strongest WINDOWS based chess program?
> >
> > 2. How strong would it be on my Pentium Pro 200?
> >
> > Thanks in advance for your answers.
> >
> > Regards,
> > James Reames
> > ches...@mcs.com
> > http://www.mcs.net/~chessman
>
> Wait for Hiarcs 6 and boot your machine up in DOS or if you must
> play in a DOS box but it will play weaker. Only tests will show how
> much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
> game match. We just need a sponsor to prove this.
OK, KK - my $10,000 says it won't.
Chris Whittington
Vincent Diepeveen
In <3311C6...@nwlink.com> brucemo <bru...@nwlink.com> writes:
>Komputer Korner wrote:
>
>> Wait for Hiarcs 6 and boot your machine up in DOS or if you must
>> play in a DOS box but it will play weaker. Only tests will show how
>> much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
>> game match. We just need a sponsor to prove this.
Why don't you just take first 6 games from a master class chessplayer,
like me? Just send me Hiarcs 6, and i'll show some problems the program
has. Besides this i'll also send few hundreds of games against Hiarcs,
as i'm busy setting up my second PC P166+, i already have auto232 player.
Vincent
>Martin!
>
>bruce
--
+----------------------------------------------------+
| Vincent Diepeveen email: vdie...@cs.ruu.nl |
| http://www.students.cs.ruu.nl/~vdiepeve/ |
+----------------------------------------------------+
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Vincent Diepeveen <vdie...@cs.ruu.nl> wrote in article
<5ev6s0$8r0$1...@krant.cs.ruu.nl>...
> In <3311C6...@nwlink.com> brucemo <bru...@nwlink.com> writes:
>
> >Komputer Korner wrote:
> >
> >> Wait for Hiarcs 6 and boot your machine up in DOS or if you must
> >> play in a DOS box but it will play weaker. Only tests will show how
> >> much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
> >> game match. We just need a sponsor to prove this.
>
> Why don't you just take first 6 games from a master class chessplayer,
> like me? Just send me Hiarcs 6, and i'll show some problems the program
> has. Besides this i'll also send few hundreds of games against Hiarcs,
> as i'm busy setting up my second PC P166+, i already have auto232 player.
>
> Vincent
> >
Why don't you just take first 6 posts from a master class dickhead,
like me? Just send me anything, and i'll show some problems you all
has. Besides this i'll also send few hundreds of posts against everybody,
as i'm busy setting up my full frontal lobotomy, i already have auto232
Louwman parrot.
Vincent "the dork" Droop
Chris Whittington
Komputer Korner
Moritz Berger wrote:
>
> On Mon, 24 Feb 1997 03:03:44 -0500, Komputer Korner <kor...@netcom.ca>
> wrote:
> <snip>
> >play in a DOS box but it will play weaker. Only tests will show how
> >much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
> >game match. We just need a sponsor to prove this.
>
> I will set up a match of a 2280 FIDE ELO player against Hiarcs 6 as
> soon as it will be released. 6 games should be ok. Time controls maybe
> 30/60, machine will be a P5/166 unless I can get hold of a Klamath
> machine soon enough ...
>
> Moritz
>
> -------------
> Moritz...@msn.com
Your 2280 player will maybe get 1 draw.
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Chris Whittington <chr...@demon.co.uk> wrote in article
<01bc236d$c5ff94c0$c308...@cpsoft.demon.co.uk>...
>
> --
> http://www.demon.co.uk/oxford-soft
>
> Vincent Diepeveen <vdie...@cs.ruu.nl> wrote in article
> <5ev6s0$8r0$1...@krant.cs.ruu.nl>...
> > In <3311C6...@nwlink.com> brucemo <bru...@nwlink.com> writes:
> >
> > >Komputer Korner wrote:
> > >
> > >> Wait for Hiarcs 6 and boot your machine up in DOS or if you must
> > >> play in a DOS box but it will play weaker. Only tests will show how
> > >> much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
> > >> game match. We just need a sponsor to prove this.
> >
> > Why don't you just take first 6 games from a master class chessplayer,
> > like me? Just send me Hiarcs 6, and i'll show some problems the program
> > has. Besides this i'll also send few hundreds of games against Hiarcs,
> > as i'm busy setting up my second PC P166+, i already have auto232
player.
> >
> > Vincent
> > >
>
> Why don't you just take first 6 posts from a master class dickhead,
> like me? Just send me anything, and i'll show some problems you all
> has. Besides this i'll also send few hundreds of posts against everybody,
> as i'm busy setting up my full frontal lobotomy, i already have auto232
> Louwman parrot.
>
> Vincent "the dork" Droop
>
>
> Chris Whittington
On reflection - this was a little over the top.
Guess I owe you a beer, Vincent :)
Komputer Korner
Chris Whittington wrote:
>
> --
snipped.
>
> OK, KK - my $10,000 says it won't.
>
> Chris Whittington
>
> >
> > --
> > Komputer Korner
> >
> > The inkompetent komputer.
> >
Your offer is pretty tempting but if the human won, it would bankrupt
me. How would we choose the IM just in case I can't resist the
temptation?
Komputer Korner
Vincent Diepeveen wrote:
>
> In <3311C6...@nwlink.com> brucemo <bru...@nwlink.com> writes:
>
> >Komputer Korner wrote:
> >
> >> Wait for Hiarcs 6 and boot your machine up in DOS or if you must
> >> play in a DOS box but it will play weaker. Only tests will show how
> >> much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
> >> game match. We just need a sponsor to prove this.
>
> Why don't you just take first 6 games from a master class chessplayer,
> like me? Just send me Hiarcs 6, and i'll show some problems the program
> has. Besides this i'll also send few hundreds of games against Hiarcs,
> as i'm busy setting up my second PC P166+, i already have auto232 player.
>
> Vincent
>
> >
> >bruce
> --
> +----------------------------------------------------+
> | Vincent Diepeveen email: vdie...@cs.ruu.nl |
> | http://www.students.cs.ruu.nl/~vdiepeve/ |
> +----------------------------------------------------+
If I sent you the beta of Hiarcs 6, I really would be guilty of
distributing copy protected software around the world as you have
accused me of about 1 month ago which I never received an apology
for BTW.
mclane
"Chris Whittington" <chr...@demon.co.uk> wrote:
>OK, KK - my $10,000 says it won't.
>Chris Whittington
Brilliant. Maybe I can convince Bernd to lose against Hiarcs and then
you give me 10.000 $ !! :-)
This is called a tuned engine !!
You can tune against anybody this way.
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
: Chris Whittington wrote:
: >
: > --
: snipped.
: > OK, KK - my $10,000 says it won't.
: >
: > Chris Whittington
: >
: > >
: > > --
: > > Komputer Korner
: > >
: > > The inkompetent komputer.
: Your offer is pretty tempting but if the human won, it would bankrupt
: me. How would we choose the IM just in case I can't resist the
: temptation?
: --
: Komputer Korner
: The inkompetent komputer.
Let me pick one and talk to him. I've been watching these guys play for
two years now. I can pick one such that if he loses, I'll face East and
bow to Mecca and repent and become a believer... :)
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
<3314D5...@netcom.ca>...
> Chris Whittington wrote:
> >
> > --
> snipped.
> >
> > OK, KK - my $10,000 says it won't.
> >
> > Chris Whittington
> >
> > >
> > > --
> > > Komputer Korner
> > >
> > > The inkompetent komputer.
> > >
>
> Your offer is pretty tempting but if the human won, it would bankrupt
> me.
It was meant to be tempting, and to make you think.
Like are you really so sure ?
Like against *any* IM ?
> How would we choose the IM just in case I can't resist the
> temptation?
There's your problem. Think what *any* IM means. If you say *any* in this
context you really mean *all*.
If a,b,c,d,e,f..... are IM's then to offer to beat *any*, you have to beat
a and b abd c and d and ......
KK, instead of making these wild claims, why don't we just wait for AEGON.
Then yuo can see how the *general* class of micros perform against some
good players. It will be tricky to point to any spectacular performance
from any one program since the game number (6) is too low.
And with several micro programs, you'll expect a spread of results from say
5 to 2.5 points out of six. Just picking the ones that score well and
ignoring the ones that don't won't tell us much.
Further, as Thorsten never tires of telling us, just adding up the game
scores is evaluating *quantity*. Looking at individual games for plans,
style, move, general capabilities is looking at *quality*.
quality assessments tell us much more than quantity ones.
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Robert Hyatt <hy...@crafty.cis.uab.edu> wrote in article
<5f337o$e...@juniper.cis.uab.edu>...
> Komputer Korner (kor...@netcom.ca) wrote:
> : Chris Whittington wrote:
> : >
> : > --
> : snipped.
> : >
> : > OK, KK - my $10,000 says it won't.
> : >
> : > Chris Whittington
> : >
> : > >
> : > > --
> : > > Komputer Korner
> : > >
> : > > The inkompetent komputer.
> : > >
>
> : Your offer is pretty tempting but if the human won, it would bankrupt
> : me. How would we choose the IM just in case I can't resist the
> : temptation?
> : --
> : Komputer Korner
>
> : The inkompetent komputer.
>
> Let me pick one and talk to him. I've been watching these guys play for
> two years now. I can pick one such that if he loses, I'll face East and
> bow to Mecca
That suggest's what I might then want to do with the $10,000 .... :)
Chris Whittington
Ed Schroder
From: "Chris Whittington" <chr...@demon.co.uk>
: Like are you really so sure ?
: Like against *any* IM ?
> How would we choose the IM just in case I can't resist the
> temptation?
: There's your problem. Think what *any* IM means. If you say *any* in
: this context you really mean *all*.
: If a,b,c,d,e,f..... are IM's then to offer to beat *any*, you have to
: beat a and b abd c and d and ......
: KK, instead of making these wild claims, why don't we just wait for
: AEGON. Then yuo can see how the *general* class of micros perform
: against some good players. It will be tricky to point to any spectacular
: performance from any one program since the game number (6) is too low.
: And with several micro programs, you'll expect a spread of results from
: say 5 to 2.5 points out of six. Just picking the ones that score well
: and ignoring the ones that don't won't tell us much.
This needs some fine tuning...
Let's take the AEGON results of the last 3 years.
Hiarcs, Rebel, Quest(=Fritz) and Mchess scored OVER 2400 ELO.
These are 3 x 4 x 6 = 72 games.
If I count the double entries of Rebel (Mephisto Advantage, Rebel7) the
result is still over ELO 2400 and the total = 84 games.
1994 : mainly 486/66 (and some P60)
1995 : mainly P90
1996 : mainly P166 (and some PP200)
: Further, as Thorsten never tires of telling us, just adding up the game
: scores is evaluating *quantity*. Looking at individual games for plans,
: style, move, general capabilities is looking at *quality*.
: quality assessments tell us much more than quantity ones.
It depends how you look at the subject!!
I always believed (and said) that:
- SSDF ratings are 50-70 to high;
- AEGON ratings are MUCH to high;
But I now am doubting...
I am doubting because the above is based on "feelings" such as:
- Computers don't play REAL chess;
- Computers ONLY win from GM's if the GM makes a blunder;
- Computers ONLY win from IM's after a few weak positional moves of an IM;
All the above are true observations in my opinion.
So I concluded, AEGON and SSDF ratings are too high!
But is it right to think this way about ELO performance?
Does ELO not have something to do with "facts" instead of "feelings"?
Sure, chess programs sometimes rape chess, they go their OWN way and it's
certainly not the human way as *we* are used to play chess, but isn't it
unfair to say "They don't play chess like US, so they are too high rated".
Isn't this an error in our superior way of thinking? :))
Isn't more fair to say "Computers simply do it in ANOTHER way"?
Look at the AEGON facts:
- over ELO 2400 in the last 3 years for Rebel, Quest, Hiarcs and Mchess;
- increasing average ELO every year; (due to hard- and software)
- increasing number of IM's and GM's every year;
- 1996 introducing the Bronstein method in favor for the humans;
Despite of this, my feelings still are saying "NO, 2400 is much to high!"
But those IRRITATING facts...
Perhaps KK is right, take 5 IM's and play 5 matches.
My "feelings" says: IM versus PC 5-0 !
But the facts (reality) is maybe 3-2 or 4-1 in favor for the PC.
Perhaps we entered a new area without even realizing it?
Just my thoughts and still doubting...:))
- Ed Schroder -
: Chris Whittington
brucemo
Robert Hyatt wrote:
>
> Komputer Korner (kor...@netcom.ca) wrote:
> : Your offer is pretty tempting but if the human won, it would bankrupt
> : me. How would we choose the IM just in case I can't resist the
> : temptation?
> Let me pick one and talk to him. I've been watching these guys play for
> two years now. I can pick one such that if he loses, I'll face East and
> bow to Mecca and repent and become a believer... :)
If you hold this match on ICC you might be able to get them to contribute some
micro-bucks.
bruce
Tom C. Kerrigan
Chris Whittington (chr...@demon.co.uk) wrote:
> Why don't you just take first 6 posts from a master class dickhead,
> like me? Just send me anything, and i'll show some problems you all
> has. Besides this i'll also send few hundreds of posts against everybody,
> as i'm busy setting up my full frontal lobotomy, i already have auto232
> Louwman parrot.
> Vincent "the dork" Droop
Oh, Chris, this is a CLASSIC!!
I didn't realize there were competitions for being a dickhead... Vincent
is only master class? There's a scary thought...
Cheers,
Tom
Robert Hyatt
brucemo (bru...@nwlink.com) wrote:
: bruce
Actually, Bruce and I talked about this via phone today. It turns out that
I had stupidly omitted one *very dangerous* IM that would be an absolutely
excellent choice for such a match. I won't give his name, but he posts here
regularly, and I notice that he "binged" a bunch of 2 0 games against Crafty
on chess.net the other night, so he's playing on a server again (haven't seen
him in about a year or so on ICC.) If he's willing, he might be one of the
most dangerous IM's around against a computer. Is the "mystery IM" willing
to step forward? :)
It would be a fun event... I'm willing to prompt Darooha to see if he would
kick in an incentive for the IM to play a match of N standard games on a
reasonable schedule... Maybe every evening, or whatever. I won't give the
IM away by mentioning timezone, but we could pick a time that would work no
matter what country he's from. And it would be illuminating...
Ed Schroder
From: "Chris Whittington" <chr...@demon.co.uk>
: Like are you really so sure ?
: Like against *any* IM ?
> How would we choose the IM just in case I can't resist the
> temptation?
: There's your problem. Think what *any* IM means. If you say *any* in
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Ed Schroder <rebc...@xs4all.nl> wrote in article
<85706724...@gln01-12.dial.xs4all.nl>...
Ed, you're cherry picking.
Add in the results of the de Konig programs in all their variants: CM5000,
the KIng etc.
Add in the results of the Genius/Mephisto programs in all their variants.
Doesn't the 'average' elo start to drop ?
If you've only a few games, and you select those programs you want, you can
get a high elo.
Chris Whittington
Komputer Korner
> : play in a DOS box but it will play weaker. Only tests will show how
> : much, but I say that Hiarcs 6 will defeat any IM in the world in a 6
> : game match. We just need a sponsor to prove this.
>
>
> KK, this is starting to move beyond your usual boundless
> enthusiasm for some new program....
"We" means the Chess Federation of Canada who are willing to have the
match take place in their offices.
Komputer Korner
Robert Hyatt wrote:
>
snipped
> It would be a fun event... I'm willing to prompt Darooha to see if he would
> kick in an incentive for the IM to play a match of N standard games on a
> reasonable schedule... Maybe every evening, or whatever. I won't give the
> IM away by mentioning timezone, but we could pick a time that would work no
> matter what country he's from. And it would be illuminating...
Bob, could you talk to Darooha about sponsoring one half of the
IM Hergott- Hiarcs 6 match or is Dean at 2390 Fide not strong
enough?
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
OK... let's fix a couple of details first...
1. location. Two possibilities. chess.net, and ICC. chess.net has
the advantage of completely free access, so that anyone can register at
no cost and observer... ICC costs, but for members that are already
there, this doesn't count. I know Roman much better and can ask him
to do this as well. However, I'll go with the majority.
2. IM. I really don't care who it is, or whether he's the best we
can find or not. If we get a really strong one, and he wins easily
we don't know any more than we do now. We could go with Hergott, or
I could also ask Roman to suggest one. Again, let's go with majority
rules.
I don't mind asking Darooha either, although I communicate regularly
with Roman via phone, and think this is exactly the kind of thing he'd
like to see/promote.
How about some opinions here. Location? Do we go for the highest
rated IM we can find, the one we think will do best against computers,
or perhaps a "middle" IM with the idea that if it is pretty close, or
if the computer wins, we can repeat with a higher-rated IM.
I'm game for anything...
Ed Schroder
From: "Chris Whittington" <chr...@demon.co.uk>
No cherry picking Chris, my point is if you take the above programs you
mention and compare the AEGON ELO results with the ones on SSDF they
simply match pretty good!
I hear the following over and over: "The rating of chess programs is
higher against other computers than against human players".
But is that true?
Isn't that only based on our opinion that humans play better chess?
I agree on this point no doubt about that... :))
Computers simply do it in a DIFFERENT way, they score which is the main
goal of the game. No?
I have not checked, but take all programs listed on SSDF who also
participated at AEGON 95/96. Calculate the AVERAGE ELO of Aegon and SSDF.
I am not sure but the average ELO of AEGON is perhaps higher than the
average ELO of SSDF... :)
Robert Hyatt
Ed Schroder (rebc...@xs4all.nl) wrote:
: No cherry picking Chris, my point is if you take the above programs you
: mention and compare the AEGON ELO results with the ones on SSDF they
: simply match pretty good!
: I hear the following over and over: "The rating of chess programs is
: higher against other computers than against human players".
: But is that true?
I don't believe that as a "blanket" statement either. Because I believe
there are some programs which perform better against humans, than they do
against programs that are tuned to only do well against other programs and
to do as well as possible in computer vs computer games...
: Isn't that only based on our opinion that humans play better chess?
: I agree on this point no doubt about that... :))
: Computers simply do it in a DIFFERENT way, they score which is the main
: goal of the game. No?
: I have not checked, but take all programs listed on SSDF who also
: participated at AEGON 95/96. Calculate the AVERAGE ELO of Aegon and SSDF.
:
: I am not sure but the average ELO of AEGON is perhaps higher than the
: average ELO of SSDF... :)
: - Ed Schroder -
Maybe the upcoming Hiarcs 6 match we're working on will shed some light. Maybe
we can do this for the top 3-4 programs over the next few months. It'd be nice
to get some real rating info. And maybe we can also get this folded in to the
SSDF list to update it a little with respect to humans. In any case, it will
be interesting to see what happens..
Ed Schroder
From: hy...@crafty.cis.uab.edu (Robert Hyatt)
: No cherry picking Chris, my point is if you take the above programs you
: mention and compare the AEGON ELO results with the ones on SSDF they
: simply match pretty good!
: I hear the following over and over: "The rating of chess programs is
: higher against other computers than against human players".
: But is that true?
> I don't believe that as a "blanket" statement either. Because I
> believe there are some programs which perform better against humans,
> than they do against programs that are tuned to only do well against
> other programs and to do as well as possible in computer vs computer
> games...
Bob, do you have proof for that?
- tune comp-comp;
- tune comp-human;
To me it's impossible to tune your chess program to perform better
against humans. It's already impossible to tune your program against
other programs. We had this discussion before. Remember Crafty 80%
against Genius.
Sure you can tune against program_X and win 20-30 elo but you may lose
30-40 elo against program_Y.
Tuning against 5,000,000 humans?
They all play different chess, ask Kasparov, Karpov, Anand etc.
Impossible job to me.
: Isn't that only based on our opinion that humans play better chess?
: I agree on this point no doubt about that... :))
: Computers simply do it in a DIFFERENT way, they score which is the main
: goal of the game. No?
: I have not checked, but take all programs listed on SSDF who also
: participated at AEGON 95/96. Calculate the AVERAGE ELO of Aegon and
: SSDF.
:
: I am not sure but the average ELO of AEGON is perhaps higher than the
: average ELO of SSDF... :)
: - Ed Schroder -
> Maybe the upcoming Hiarcs 6 match we're working on will shed some light.
> Maybe we can do this for the top 3-4 programs over the next few months.
> It'd be nice to get some real rating info. And maybe we can also get
> this folded in to the SSDF list to update it a little with respect to
> humans. In any case, it will be interesting to see what happens..
Yes.
- Ed Schroder -
Robert Hyatt
Ed Schroder (rebc...@xs4all.nl) wrote:
: From: hy...@crafty.cis.uab.edu (Robert Hyatt)
: : No cherry picking Chris, my point is if you take the above programs you
: : mention and compare the AEGON ELO results with the ones on SSDF they
: : simply match pretty good!
: : I hear the following over and over: "The rating of chess programs is
: : higher against other computers than against human players".
: : But is that true?
: > I don't believe that as a "blanket" statement either. Because I
: > believe there are some programs which perform better against humans,
: > than they do against programs that are tuned to only do well against
: > other programs and to do as well as possible in computer vs computer
: > games...
: Bob, do you have proof for that?
: - tune comp-comp;
: - tune comp-human;
Yes. And here is the way to confirm it. Don't do what Bruce and I
are doing with asymmetric king safety. Play on ICC for a few months.
I'll bet that Crafty/Ferret stay at least 200 points higher than you
do with your present eval. Why? Because there are some masters and
IM players there that are simply going to rip your head off. I've had
it happen far too many times. Against one of them I'd much rather
sacrifice a pawn than open files around my king. Bruce and I discussed
this, modified the king safety, and iterated over and over. Against
humans, Crafty's king safety can easily get to -3 with material even.
I also have code in crafty that turns this off against computers, because
computers don't attack like humans. So that king safety is not nearly so
important as material, generally. Humans are also quite a bit more
"relaxed" about a pawn, so that they are often willing to sac a pawn, or
an exchange, to get a position that is nearly impossible to lose, and where
they might win. Meanwhile the program will be trolling along at +2. I've
modified eval terms in crafty to the point where I have seen +4 with material
even, and I've seen -4 for the same reason. And it doesn't work particularly
well against programs that don't overlook anything tactically.
ICC's a great learning institution, because you'll see that playing 10
games against a GM in a year, is *not* the same thing as playing 100
games against the same GM in a week. You have a hole, he'll send a rocket
through it. over and over and over. Roman is right now continually
complaining about Crafty allowing the classic stonewall position to come
up. You wouldn't believe the different ways he finds to transpose to these
positions. However, against a good human, no computer is going to survive
them either.
So yes, tuning against humans is different than tuning against computers.
Crafty "adapts" and adjusts some things depending on whether it is playing
a computer or not, which is wny the "name xxx" command is critical for
computer vs computer games, and why the "computer list" inside crafty is
also important.
I could show some games played by crafty on chess.net where it was getting
zapped by genius over and over, because it didn't know this particular
program was genius. I fixed the "list" and it began winning consistently
again. Over 5 days, it's rating went from 2600 (chess.net ratings are quite
a bit lower than ICC ratings) to just under 2500. I added the names of the
new computers there to my list so crafty could adjust its scoring and get
rid of the big king safety stuff, and it's now back to over 2600 again. With
no changes other than letting it know who's a computer and who's not. This
is obviously not 100% convincing that this works, but it's a strong indicator
to me...
: To me it's impossible to tune your chess program to perform better
: against humans. It's already impossible to tune your program against
: other programs. We had this discussion before. Remember Crafty 80%
: against Genius.
: Sure you can tune against program_X and win 20-30 elo but you may lose
: 30-40 elo against program_Y.
: Tuning against 5,000,000 humans?
: They all play different chess, ask Kasparov, Karpov, Anand etc.
: Impossible job to me.
I realize that it looks that way. But against computers, they don't all
play different chess. That's the *big* key. The good players know that
most programs simply don't understand long-range threats against their
kings until they see something happen tactically. If you know that, most
games are over before they start. Again, you should try "greg1", or
"dreamteamer", or "schroer", or "tim", or any of a dozen other masters,
IM's and even GM's. Until we close this huge hole, they are going to keep
shooting rockets through it. Once we fix this, then we see more real chess
and you will be closer to the truth.
If you go back about 15 years, there was a well-known and trivial way to
beat all the programs... Just trade pieces and win the ending, where they
were horrible. We fixed that. Now we have a more difficult problem to fix,
king-side safety. Just counting pawns and looking at where they are is *not*
enough, from experience. There's oh so juch more to it, and we aren't doing
it very well. I believe the best of the best at this right now is probably
Kittinger's WchessX program. But even it is not good enough yet...
: : Isn't that only based on our opinion that humans play better chess?
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Ed Schroder <rebc...@xs4all.nl> wrote in article
<5f8q6t$gvv$2...@news1.xs4all.nl>...
> No cherry picking Chris, my point is if you take the above programs you
> mention and compare the AEGON ELO results with the ones on SSDF they
> simply match pretty good!
>
> I hear the following over and over: "The rating of chess programs is
> higher against other computers than against human players".
>
> But is that true?
>
> Isn't that only based on our opinion that humans play better chess?
> I agree on this point no doubt about that... :))
>
> Computers simply do it in a DIFFERENT way, they score which is the main
> goal of the game. No?
>
> I have not checked, but take all programs listed on SSDF who also
> participated at AEGON 95/96. Calculate the AVERAGE ELO of Aegon and SSDF.
>
> I am not sure but the average ELO of AEGON is perhaps higher than the
> average ELO of SSDF... :)
No, I've not checked by doing the calculation. Lets do it after the next
AEGON, for all SSDF rated programs.
What I do notive at AEGON is that the IMs and GMs fall into two classes.
Those that have anti-computer experience.
And those that don't.
The 'old hands' win (usually), the new ones, at AEGON for the first time
are trying their way, and often lose.
Van der Weil, for example, raraley loses a game. He has insights on this.
Famous Van der Weil saying (paraphrased): "they (the programs) don't play
chess, but their own subset of the game. As soon as you realise this, you
just play to counter their way, its easy."
Interesting would be stats on IM, GM improvement after playing 6,12,18
games and so on.
I think we are all going to have some beers together in April :)
One complaint about AEGON: they hold it at the same time the tulips bloom.
The pollen count is through the roof. If you get hay fever, even just
occasionally, you die at AEGON :(
Chris Whittington
brucemo
Robert Hyatt wrote:
> Yes. And here is the way to confirm it. Don't do what Bruce and I
> are doing with asymmetric king safety. Play on ICC for a few months.
> I'll bet that Crafty/Ferret stay at least 200 points higher than you
> do with your present eval. Why? Because there are some masters and
> IM players there that are simply going to rip your head off. I've had
> it happen far too many times. Against one of them I'd much rather
> sacrifice a pawn than open files around my king. Bruce and I discussed
> this, modified the king safety, and iterated over and over. Against
> humans, Crafty's king safety can easily get to -3 with material even.
I would love to see any of the professionals put an automatic program on ICC,
but I don't agree that ours would be 200 points higher than any of them,
specifically Rebel.
I think there's a good chance he'd hit the top of the list, at least for a
while.
Ratings are random anyway, and for all we know his king safety may be better.
I don't think that my king safety is all that great. It's just paranoid. And
even at that, the program still gets mated a lot. And my opening book sucks,
so the thing allows people to get positions where the program can barely
develop its pieces, much less avoid getting mated.
And in any case, probably a huge number of the games he'd get there would be
other computers :-)
bruce
brucemo
Chris Whittington wrote:
> One complaint about AEGON: they hold it at the same time the tulips bloom.
> The pollen count is through the roof. If you get hay fever, even just
> occasionally, you die at AEGON :(
This is GOOD news for us. I don't know about you, but I can operate just
fine when my nose is stuffed, I had a perfectly violent cold in Jakarta. I
doubt the GM's play chess as well in the same condition :-)
bruce
Ed Schroder
From: hy...@crafty.cis.uab.edu (Robert Hyatt)
: : I hear the following over and over: "The rating of chess programs is
: : higher against other computers than against human players".
: : But is that true?
: > I don't believe that as a "blanket" statement either. Because I
: > believe there are some programs which perform better against humans,
: > than they do against programs that are tuned to only do well against
: > other programs and to do as well as possible in computer vs computer
: > games...
: Bob, do you have proof for that?
: - tune comp-comp;
: - tune comp-human;
> Yes. And here is the way to confirm it. Don't do what Bruce and I
> are doing with asymmetric king safety. Play on ICC for a few months.
> I'll bet that Crafty/Ferret stay at least 200 points higher than you
> do with your present eval. Why? Because there are some masters and
> IM players there that are simply going to rip your head off. I've had
> it happen far too many times. Against one of them I'd much rather
> sacrifice a pawn than open files around my king. Bruce and I discussed
> this, modified the king safety, and iterated over and over. Against
> humans, Crafty's king safety can easily get to -3 with material even.
But isn't that just normal chess knowledge every chess program used to
have?
I checked Rebel, the king safety goes up from -5.75 to +5.75 as minimum
and maximum. No asymmetry yet... :)
> I also have code in crafty that turns this off against computers,
> because computers don't attack like humans. So that king safety is
> not nearly so important as material, generally. Humans are also quite
> a bit more "relaxed" about a pawn, so that they are often willing to
> sac a pawn, or an exchange, to get a position that is nearly impossible
> to lose, and where they might win. Meanwhile the program will be
> trolling along at +2. I've modified eval terms in crafty to the point
> where I have seen +4 with material even, and I've seen -4 for the same
> reason. And it doesn't work particularly well against programs that
> don't overlook anything tactically.
I suppose your "king safety" feature is set off in the upcoming
challenge then... :))
Remember I just informed about my -5.75 to +5.75 king safety window so
you may change your mind accordingly... :))
> ICC's a great learning institution, because you'll see that playing 10
> games against a GM in a year, is *not* the same thing as playing 100
> games against the same GM in a week. You have a hole, he'll send a
> rocket through it. over and over and over. Roman is right now
> continually complaining about Crafty allowing the classic stonewall
> position to come up. You wouldn't believe the different ways he finds
> to transpose to these positions. However, against a good human, no
> computer is going to survive them either.
I agree, on Aegon I see the same. It's impressive to see how GM's are
setting up an attack with deep positional knowledge. It reminds me
(despite of the high elo of the computers) how poor chess programs play
human chess. It's certainly more fun to see how how a GM kills a computer
than the other way around.
> So yes, tuning against humans is different than tuning against
> computers. Crafty "adapts" and adjusts some things depending on
> whether it is playing a computer or not, which is wny the "name xxx"
> command is critical for computer vs computer games, and why the
> "computer list" inside crafty is also important.
But that's no tuning because YOU decide who the opponent is! If Crafty
(so the software) is able to recognize the human or computer I think you
do have a point.
> I could show some games played by crafty on chess.net where it was
> getting zapped by genius over and over, because it didn't know this
> particular program was genius. I fixed the "list" and it began
> winning consistently again. Over 5 days, it's rating went from 2600
> (chess.net ratings are quite a bit lower than ICC ratings) to just
> under 2500. I added the names of the new computers there to my list
> so crafty could adjust its scoring and get rid of the big king safety
> stuff, and it's now back to over 2600 again. With no changes other
> than letting it know who's a computer and who's not. This is
> obviously not 100% convincing that this works, but it's a strong
> indicator to me...
I remeber our discussion about "knowing the opponent" is at least worth
100 ELO. I tend to agree on this.
But what you are doing with Crafty is that you (or somebody else) decides
in favor of Crafty who the opponent is. IMO this is not tuning since the
human still decides and not the software.
I know Rebel scores better against Genius if I change the playing style
of Rebel! I know Genius scores better against Rebel if I change the
playing style of Genius!
But the point is does Rebel or Genius know this by software?
Now that is tuning and would make chess programs a lot stronger.
Rebel does not contain software to recognize the opponent.
Wish I was able... :))
[ snip ]
- Ed Schroder -
mclane
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>: : But is that true?
Sorry Bob. It is very difficult to show evidence because Chris program
is - in the moment not at its best - but Chris program KNOWS about
King-attack and ALSO about king-defence.
The same way it attacks (the older versions) like a maniac, it sees
attacks.
Of course it cannot play accurate. That is one main problem. CSTal
cannot play accurate because it is not taught to do this. It is not
designed for accuracy. It can only bluff. But first you have to show
it, that it is wrong. That is not very easy....
AGAIN: I refer to the BETTER versions I have in my big CSTAL
archive/collection.
>I also have code in crafty that turns this off against computers, because
>computers don't attack like humans. So that king safety is not nearly so
>important as material, generally. Humans are also quite a bit more
>"relaxed" about a pawn, so that they are often willing to sac a pawn, or
>an exchange, to get a position that is nearly impossible to lose, and where
>they might win. Meanwhile the program will be trolling along at +2. I've
>modified eval terms in crafty to the point where I have seen +4 with material
>even, and I've seen -4 for the same reason. And it doesn't work particularly
>well against programs that don't overlook anything tactically.
Good idea. Still I am believing in a UNIVERSAL program that plays good
against humans AND computers and both in the same style. But this is
maybe a DREAM of childhood....
>ICC's a great learning institution, because you'll see that playing 10
>games against a GM in a year, is *not* the same thing as playing 100
>games against the same GM in a week. You have a hole, he'll send a rocket
>through it. over and over and over. Roman is right now continually
>complaining about Crafty allowing the classic stonewall position to come
>up. You wouldn't believe the different ways he finds to transpose to these
>positions. However, against a good human, no computer is going to survive
>them either.
Right.
>So yes, tuning against humans is different than tuning against computers.
Right. And you have done it. I was really able to feel it, right from
my first test-game. THIS was the reason I - who am normally against
ALL fast/stupid programs (I do not imply that yours is INTO this group
anyway) was surprised about your program !!
It played human although it is designed to be fast and with simple but
clever and smart evaluations.
If this is the way, I have nothing against it. I can live with it. I
can also use YOUR program for mail-chess. I have more problems with
Fritz4. But you can call my problems subjective and not important for
other people.
>Crafty "adapts" and adjusts some things depending on whether it is playing
>a computer or not, which is wny the "name xxx" command is critical for
>computer vs computer games, and why the "computer list" inside crafty is
>also important.
>I could show some games played by crafty on chess.net where it was getting
>zapped by genius over and over, because it didn't know this particular
>program was genius. I fixed the "list" and it began winning consistently
>again. Over 5 days, it's rating went from 2600 (chess.net ratings are quite
>a bit lower than ICC ratings) to just under 2500. I added the names of the
>new computers there to my list so crafty could adjust its scoring and get
>rid of the big king safety stuff, and it's now back to over 2600 again. With
>no changes other than letting it know who's a computer and who's not.
Of course you have understood and managed an important fact:
If somebody knows the opponent, he can magically change the result.
This is worth 50 to 80 ELO ?!
Maybe more.
>This
>is obviously not 100% convincing that this works, but it's a strong indicator
>to me...
I believe you.
>: To me it's impossible to tune your chess program to perform better
>: against humans. It's already impossible to tune your program against
>: other programs. We had this discussion before. Remember Crafty 80%
>: against Genius.
>: Sure you can tune against program_X and win 20-30 elo but you may lose
>: 30-40 elo against program_Y.
>: Tuning against 5,000,000 humans?
>: They all play different chess, ask Kasparov, Karpov, Anand etc.
>: Impossible job to me.
>I realize that it looks that way. But against computers, they don't all
>play different chess.
Oh yes - exactly what I said a few minutes ago. Many chess-programs
are THE SAME.
>That's the *big* key. The good players know that
>most programs simply don't understand long-range threats against their
>kings until they see something happen tactically. If you know that, most
>games are over before they start.
Brilliant analysis. Bob - I hope german telekom monopol will be ended
early that WE germans can participate and talk with you guys on the
ICC servers without spending all our money for telephone costs.
I would really like to chat with you while watching a nice game,
because when I read the depth of your analysis concerning games
between computers and humans, I can see that we have same conclusions,
although you have seen more computer vs. humans game on ICC.
> Again, you should try "greg1", or
>"dreamteamer", or "schroer", or "tim", or any of a dozen other masters,
>IM's and even GM's. Until we close this huge hole, they are going to keep
>shooting rockets through it. Once we fix this, then we see more real chess
>and you will be closer to the truth.
>If you go back about 15 years, there was a well-known and trivial way to
>beat all the programs... Just trade pieces and win the ending, where they
>were horrible. We fixed that. Now we have a more difficult problem to fix,
>king-side safety. Just counting pawns and looking at where they are is *not*
>enough, from experience. There's oh so juch more to it, and we aren't doing
>it very well. I believe the best of the best at this right now is probably
>Kittinger's WchessX program. But even it is not good enough yet...
Yes. Kittingers Wchess has somehow fixed this. Even for Chris' program
it was very difficult to win against Wchess. But some versions were
able to king-attack Wchess.
When will Dave give us another Wchess that makes a step forward.
DAVE ? Can you hear us ? We/I want a new , 3rd superconny remake !!!
mclane
"Chris Whittington" <chr...@demon.co.uk> wrote:
>No, I've not checked by doing the calculation. Lets do it after the next
>AEGON, for all SSDF rated programs.
>What I do notive at AEGON is that the IMs and GMs fall into two classes.
>Those that have anti-computer experience.
>And those that don't.
>The 'old hands' win (usually), the new ones, at AEGON for the first time
>are trying their way, and often lose.
>Van der Weil, for example, raraley loses a game. He has insights on this.
>Famous Van der Weil saying (paraphrased): "they (the programs) don't play
>chess, but their own subset of the game. As soon as you realise this, you
>just play to counter their way, its easy."
They emulate chess !!! BTW: wasn't one of the programs of YOU called
Chess-Simulator ?! A very precise name, isn't it ?
>Interesting would be stats on IM, GM improvement after playing 6,12,18
>games and so on.
>I think we are all going to have some beers together in April :)
Give me a change for a civilized counter-attack drinking my tea. I
have always to sit barbarish between you eating cheese, smoking
cigarettes (Hello ED), drinking beer (Hello Peter Schreiner/Gillgasch
and and and ), and I would like to come into the same bloomy mood, but
getting a cup of tea, maybe the british way.... like chris always
explains me how to do it right, is not easy. So I dy thirsty, hungry,
and in vulcan-mood meanwhile you sing and search the hotal key as if
you were blind ...
>One complaint about AEGON: they hold it at the same time the tulips bloom.
>The pollen count is through the roof. If you get hay fever, even just
>occasionally, you die at AEGON :(
If you were from Birmingham, you would not have any problems with
pollen ! As I have learned here, the people from birmingham use their
nose for many things....
mclane
>Bob, do you have proof for that?
>- tune comp-comp;
>- tune comp-human;
>To me it's impossible to tune your chess program to perform better
>against humans. It's already impossible to tune your program against
>other programs. We had this discussion before. Remember Crafty 80%
>against Genius.
>Sure you can tune against program_X and win 20-30 elo but you may lose
>30-40 elo against program_Y.
>Tuning against 5,000,000 humans?
>They all play different chess, ask Kasparov, Karpov, Anand etc.
>Impossible job to me.
No - I am (and this is an exception) one time on Bob's side.
You can TUNE or maybe luckily write a program (if you only have one
effort and it works well) that plays better against humans.
In my opinon there are some programs that play really better against
humans, than against computers.
Wchess e.g. is such a program.
Or Socrates.
Or maybe crafty.
But Ed - in the end, I know what you mentioned is also true:
You could write a program that would win against Kasparov (e.g.
Fritz), but this program would not be able to kill Karpov in a
blitz-game or even in an active-game.
Because Karpov plays very difficult.
But, and here my main point, IS THAT AN ARGUMENT F O R YOU ?
Because it is the same in human-chess.
I can play very good against Mr.x, but fail against Mr.y.
So humans have the same problems than programs have, playing against
humans.
Since we have many programs on the market the are THE SAME , we will
be able to write killer-engines that play marvellous against this type
of programs. Many of the state-of-the-art-programs have the same type
of WORKING in speed and algorithms.
So it should work to find a way to cheat against them, by writing a
program that plays different.
In the moment we have more programs that work different (when will
this be ?) you cannot write killer-engines.
But - in the moment there is much INCEST between the programs.
One copies from the other. Sometimes programmer is forced to do this,
to be state-of-the-art too. Look:
Gandalf was a very different program.
It made 1000 nodes per second on my 486-33.
That is not much.
Now Gandalf is changed into another kind of program, because it has
maybe a more normal search then before Paderborn.
Maybe Steen and Dan have tried this IDEA of using a more succesful
effort in using normal search because they saw Gandalf in prison
arround 2000 ELO. What a pity, they thought. We have so much
knowledge, and we are unable to climb above the 2000 barrier.
In my opinon NIMZO made the same development from slow/knowledged into
fast/plain (=not knowledged or even pre-processing).
This gave Chrilly Donninger and his team arround 250 ELO points, or
even more. But was it a step into right direction ? Or was it one
step, and they will be able to do the next step. Or was it a
dead-end-street ?!
We have all different opinon on this topic.
Indeed, some people have very pragmatical point of views to this, and
I can understand them. It is one thing to be a dreamer, and another
one to feed a family. Or to lead a company of employees.
There is no golden rule, no unique recepy.
Therefore I like computerchess. Although I always attack the
fast/stupid programs, I like the competition between them and the
slow/knowledged ones. E.g. in Paderborn 1997 IMO the fast/stupid ones
were able to overtake the others.
Despite this, it was a very narrow competition.
Jakarta was IMO dominated by the intelligent ones.
What will the next championship bring to us ?
>: Isn't that only based on our opinion that humans play better chess?
>: I agree on this point no doubt about that... :))
>: Computers simply do it in a DIFFERENT way, they score which is the main
>: goal of the game. No?
>: I have not checked, but take all programs listed on SSDF who also
>: participated at AEGON 95/96. Calculate the AVERAGE ELO of Aegon and
>: SSDF.
>:
>: I am not sure but the average ELO of AEGON is perhaps higher than the
>: average ELO of SSDF... :)
>: - Ed Schroder -
>> Maybe the upcoming Hiarcs 6 match we're working on will shed some light.
>> Maybe we can do this for the top 3-4 programs over the next few months.
>> It'd be nice to get some real rating info. And maybe we can also get
>> this folded in to the SSDF list to update it a little with respect to
>> humans. In any case, it will be interesting to see what happens..
IM is not IM. E.g. if I let Hiarcs6 play vs. IM Bernd Kohlweyer (who
is living and behaving like Tal, but he has a chess-playing-style like
Petrosian) or I let Hiarcs6 play against IM Markus Schaefer (who is
behaving like Petrosian in life and plays like a murderer and Tal'ish
maniac on the chess-board) the outcome could be totally different.
But - as I have said - this problem occurs the same way between
humans. I don't know much about this IM Hergott.
I know Kohlweyer and Schaefer very good. Does anybody know details
about playing-style of Hergott?
Robert Hyatt
: > Yes. And here is the way to confirm it. Don't do what Bruce and I
: > are doing with asymmetric king safety. Play on ICC for a few months.
: > I'll bet that Crafty/Ferret stay at least 200 points higher than you
: > do with your present eval. Why? Because there are some masters and
: > IM players there that are simply going to rip your head off. I've had
: > it happen far too many times. Against one of them I'd much rather
: > sacrifice a pawn than open files around my king. Bruce and I discussed
: > this, modified the king safety, and iterated over and over. Against
: > humans, Crafty's king safety can easily get to -3 with material even.
: I would love to see any of the professionals put an automatic program on ICC,
: but I don't agree that ours would be 200 points higher than any of them,
: specifically Rebel.
: I think there's a good chance he'd hit the top of the list, at least for a
: while.
"while" being the key word here. And then along comes the following players:
ADOLF, greg1, blunite, Tim, dreamteamer, beetle, <etc>... and we'd see a
feast of sorts. Rebel plays great chess. But it hasn't been exposed to
exactly this facet of chess yet. If you can't prevent the Stonewall
formation as black, or else know exactly how to defend once white reaches
such positions, it'll be ugly. I see it happen against every program that
runs on ICC, which includes all commercial programs. Just that the operator
will recognize he's getting rolled and refuses to play that particular ICC
player any more, which is one way to avoid the problem. Going "automatic"
exposes every weakness you have to the light of day...
I've personally seen this problem with Crafty, and have spent who knows how
much time trying to solve it. And I still get a "did it again" from Roman
every few days. Cooking the book isn't enough. The program has got to
understand those positions (Stonewall is but one well-known example of
course.) It's a battle of wits, with me on one side, a bunch of IMs and GMs
on the other, and Crafty right square in the middle. I've done "better"
in places, but it's a weakness that would prevent any program from having
a chance against a Kasparov at present. There are some classic patterns that
I recognize instantly, but which I don't quite know how to implement in the
eval just yet. Unfortunately, the humans don't have such a problem for the
most part.
: Ratings are random anyway, and for all we know his king safety may be better.
: I don't think that my king safety is all that great. It's just paranoid. And
: even at that, the program still gets mated a lot. And my opening book sucks,
: so the thing allows people to get positions where the program can barely
: develop its pieces, much less avoid getting mated.
: And in any case, probably a huge number of the games he'd get there would be
: other computers :-)
Yes. which is another problem, because it's probably necessary to play a
little differently against them. And eventually, as programs learn how to
attack, it's going to be necessary to play chess the right way, rather than
fudging about what you do vs carbon and vs silicon, which is just a crutch
right now. :)
: bruce
Robert Hyatt
mclane (mcl...@prima.ruhr.de) wrote:
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
: >: : But is that true?
Nothing to be sorry about. I wasn't really talking about CSTal, which
might be just what is needed to counter the humans on ICC and in real
life. Would like to see it there some to evaluate what he's done as
it would put him right alongside most of the silicon compatition. If
it does well against humans, it'd also be interesting to then let it
play most of the programs there and see how it stacks up against them.
I hope what he's doing works. Because this is a real problem for
most...
: The same way it attacks (the older versions) like a maniac, it sees
Robert Hyatt
Ed Schroder (rebc...@xs4all.nl) wrote:
: From: hy...@crafty.cis.uab.edu (Robert Hyatt)
: : : I hear the following over and over: "The rating of chess programs is
: : : higher against other computers than against human players".
: : : But is that true?
: : > I don't believe that as a "blanket" statement either. Because I
: : > believe there are some programs which perform better against humans,
: : > than they do against programs that are tuned to only do well against
: : > other programs and to do as well as possible in computer vs computer
: : > games...
: : Bob, do you have proof for that?
: : - tune comp-comp;
: : - tune comp-human;
: > Yes. And here is the way to confirm it. Don't do what Bruce and I
: > are doing with asymmetric king safety. Play on ICC for a few months.
: > I'll bet that Crafty/Ferret stay at least 200 points higher than you
: > do with your present eval. Why? Because there are some masters and
: > IM players there that are simply going to rip your head off. I've had
: > it happen far too many times. Against one of them I'd much rather
: > sacrifice a pawn than open files around my king. Bruce and I discussed
: > this, modified the king safety, and iterated over and over. Against
: > humans, Crafty's king safety can easily get to -3 with material even.
: But isn't that just normal chess knowledge every chess program used to
: have?
: I checked Rebel, the king safety goes up from -5.75 to +5.75 as minimum
: and maximum. No asymmetry yet... :)
The question that ICC would answer is this: Every think done when both
sides castle to the same side is almost "neutral" in your eval. I push a
pawn to open a file, the file is opened on both kings and is therefore
offsetting. *if* you can attack thru that file, and do so, then you don't
need asymmetry. If you aren't so good at attacking, you need it badly.
Most programs that I know of simply don't know how to evaluate how good an
attack is going to be. pieces close to the king is only a hint, but what
pieces are bearing from a distance, and on which squares, and how easy is it
to defend those squares, and so forth. Humans do this well. If you don't
(and crafty doesn't as of yet), then you can't stand that "offsetting" eval
as you will be opening up lines to your own king and to the opponent's king
without knowing if you can use them or not. And if you can't, guess who
will? :)
: > I also have code in crafty that turns this off against computers,
: > because computers don't attack like humans. So that king safety is
: > not nearly so important as material, generally. Humans are also quite
: > a bit more "relaxed" about a pawn, so that they are often willing to
: > sac a pawn, or an exchange, to get a position that is nearly impossible
: > to lose, and where they might win. Meanwhile the program will be
: > trolling along at +2. I've modified eval terms in crafty to the point
: > where I have seen +4 with material even, and I've seen -4 for the same
: > reason. And it doesn't work particularly well against programs that
: > don't overlook anything tactically.
: I suppose your "king safety" feature is set off in the upcoming
: challenge then... :))
yes. :)
: Remember I just informed about my -5.75 to +5.75 king safety window so
: you may change your mind accordingly... :))
No. I haven't seen Rebel notice how well bishops and rooks and queens
are controlling squares on the kingside yet. My king safety can reach a
max of +/-5.0 or so as well, but it's rare to see a kingside that shredded
unless the score is down arount -Mate already... :) When we can all notice
long-range attacks from sliding pieces, and figure out if the opponent has
one key square that we can "win", then we'll all become dangerous. It's a
goal of mine. It's obviously a goal of Chris's... I suspect it's a goal
of everyones...
: > ICC's a great learning institution, because you'll see that playing 10
: > games against a GM in a year, is *not* the same thing as playing 100
: > games against the same GM in a week. You have a hole, he'll send a
: > rocket through it. over and over and over. Roman is right now
: > continually complaining about Crafty allowing the classic stonewall
: > position to come up. You wouldn't believe the different ways he finds
: > to transpose to these positions. However, against a good human, no
: > computer is going to survive them either.
: I agree, on Aegon I see the same. It's impressive to see how GM's are
: setting up an attack with deep positional knowledge. It reminds me
: (despite of the high elo of the computers) how poor chess programs play
: human chess. It's certainly more fun to see how how a GM kills a computer
: than the other way around.
yes, and it's also fun to watch a game that is not about the kings, but is
about pieces defending pieces which are defending pawns. The GM normally
gets lost when the tactics go crazy. Which is why I think many of us are
so successful against GM players. Most that I know and talk with simply
refuse to play "anti-computer" for a good reason: it doesn't help them in
any way prepare for the next tournament, since computers won't be present.
As a result, they play normal chess, let tactics get a little out of hand,
and lose. Were we serious competition in (say) the interzonals and major
tournaments, they'd learn anti-computer strategies pretty quick and this
would change... There *are* exceptions of course. Roman can roll the best
of us many times. Or, if he's getting ready for a tournament, he'll get
rolled because he wants to play what he'd call normal chess...
: > So yes, tuning against humans is different than tuning against
: > computers. Crafty "adapts" and adjusts some things depending on
: > whether it is playing a computer or not, which is wny the "name xxx"
: > command is critical for computer vs computer games, and why the
: > "computer list" inside crafty is also important.
: But that's no tuning because YOU decide who the opponent is! If Crafty
: (so the software) is able to recognize the human or computer I think you
: do have a point.
Crafty does recognize this on a server. Xboard tells crafty who it is playing,
and crafty looks this up in a database to see who/what it is, and adjusts itself
accordingly. The better solution would be to keep a database that Crafty uses
to remember things about how this player plays the game....
: > I could show some games played by crafty on chess.net where it was
: > getting zapped by genius over and over, because it didn't know this
: > particular program was genius. I fixed the "list" and it began
: > winning consistently again. Over 5 days, it's rating went from 2600
: > (chess.net ratings are quite a bit lower than ICC ratings) to just
: > under 2500. I added the names of the new computers there to my list
: > so crafty could adjust its scoring and get rid of the big king safety
: > stuff, and it's now back to over 2600 again. With no changes other
: > than letting it know who's a computer and who's not. This is
: > obviously not 100% convincing that this works, but it's a strong
: > indicator to me...
: I remeber our discussion about "knowing the opponent" is at least worth
: 100 ELO. I tend to agree on this.
One of the *bad* things was that some of these opponents that were killing crafty
were crafty clones. on somewhat slower machines. But they knew crafty ws a
computer, thanks to me, but crafty didn't know about them. :)
: But what you are doing with Crafty is that you (or somebody else) decides
: in favor of Crafty who the opponent is. IMO this is not tuning since the
: human still decides and not the software.
in manual games you are correct, of course, it it won't work. Since most of
my tuning is on ICC, it does work ok since I know who my opponent is in each
game...
: I know Rebel scores better against Genius if I change the playing style
: of Rebel! I know Genius scores better against Rebel if I change the
: playing style of Genius!
: But the point is does Rebel or Genius know this by software?
: Now that is tuning and would make chess programs a lot stronger.
: Rebel does not contain software to recognize the opponent.
: Wish I was able... :))
Play on ICC. At least there, you can. :)
: [ snip ]
: - Ed Schroder -
Ingo Althoefer
Ed Schroder <rebc...@xs4all.nl> wrote:
>From: hy...@crafty.cis.uab.edu (Robert Hyatt)
>> I also have code in crafty that turns this off against computers,
>> because computers don't attack like humans. ......
>> So yes, tuning against humans is different than tuning against
>> computers. Crafty "adapts" and adjusts some things depending on
>> whether it is playing a computer or not, which is why the "name xxx"
>> command is critical for computer vs computer games, and why the
>> "computer list" inside crafty is also important .... With no changes other
>> than letting it know who's a computer and who's not. This is
I like this idea very much. And of course I would also like refinements of it.
>> obviously not 100% convincing that this works, but it's a strong
>> indicator to me...
>I remeber our discussion about "knowing the opponent" is at least worth
>100 ELO. I tend to agree on this. ...
>I know Rebel scores better against Genius if I change the playing style
>of Rebel! I know Genius scores better against Rebel if I change the
>playing style of Genius! But the point is does Rebel or Genius know this by
>software? Now that is tuning and would make chess programs a lot stronger.
>Rebel does not contain software to recognize the opponent...
In November 96 Ed Schroeder had asked a very interesting, hypothetical ques-
tion to the SSDF: "Would you allow a program to ask for the identity of its
opponents before the games ?" Unfortunately Goran Grottling overreacted,
threw Ed Schroeder in one pot with some other guy who got on his nerves,
and stopped contributing to our newsgroup. ( What about a return ? )
1. A program with an identification mechanism might perform very well in the
SSDF world (and probably also in other niches of our universe): Assume you have
two different engines E1 and E2 of same average strength. Against some oppo-
nents E1 does better, against others E2. When you have to play a 20 games
match against another program A you do the following. In the first half of
the match each of your engines plays half of the games, so 5 times E1-A and
5 times E2-A. After this you control the intermediate balances ( this must be
the right expression according to my dictionary; what I mean is "the scores" ).
If E1 made more points, you use E1 in the other 10 games. If E2 made more
points, you let E2 play the other 10 games. ( Of course you are free to compose
minor or major variants of this basic scheme. ) This computer E with two
engines E1 and E2 and a simple adder should perform better then each Ei
alone. ( E1 and E2 may differ only by a few parameters in the evaluation
functions. )
2. If you were allowed to make more complicated constructions, the following
would yield an SSDF performance of about 2560: Build a very big program BIG
including all "traditional" top programs ( Rebell 8, Hiarcs 5.6, M-Chess 6.0,
Genius 3, ...) as engines. Then also feed the current SSDF-list in this
program. Not only the ratings, but also the scores of the single matches
between "traditional" programs. Now, if BIG has to play against some program
X, look in your data which of the engines in BIG has the best score against
X. Then play with this program.
Taking into account only games against the top 25 programs in the SSDF list,
BIG would need only five engines ( = five top programs ) to get a 2560 per-
formance. However, adding more programs to BIG would not yield a much better
rating.
I do not propose to realize idea 2. But No. 1 is not this bad, or ?
Ingo Althoefer.
someone
>In article <5f337o$e...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu (Robert
>
>
>Let me pick one and talk to him. I've been watching these guys play for
>two years now. I can pick one such that if he loses, I'll face East and
>bow to Mecca and repent and become a believer... :)
>
>
Does that include the circumcision bit? :)
Andy
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Ingo Althoefer <alth...@pdec01.uucp> wrote in article
<5fcf8j$h...@fsuj19.rz.uni-jena.de>...
But do the elo improvements in either of these versions mean anything ?
On the one side chess knowledge, whether from search or evaluation in the
case of programs, or human knowledge in the case of humans, will correlate
somehow with an elo increase.
Know doubled pawn is bad equates to play stronger chess, equates to higher
elo points.
But knowing your opponent ?
Ok, so this approach may earn more elo points on the ssdf list. But then
the Mr Big, and E1 and E2 enter a tourney, separately, each has to play a
game against a human, and then another human, and then another and so on.
Why should this Mr Big play with any more chess skill than E1 or E2. And
why should Mr Big be more likely to gain more tourney points ?
Answer is Mr Big wouldn't.
Except he's got a higher SSDF grade. So what does this higher SSDF grade
mean ?
It means that Ingo has found a way (amongst all the other ways already
found) to increase performance on this list because Ingo knows how the list
is constructed, and how the SSDF make their tests.
Or look at it another way. Suppose the SSDF choose Mr Big's computer
opponents by random, no long sequences against the same opponent. The
choosing mechanism between E1 and E2 would not now work.
So Mr Big performs on the SSDF in relation to how the SSDF order the games
:)
Chris Whittington
>
> Ingo Althoefer.
>
Komputer Korner
mclane wrote:
>
snipped
>
> >: - Ed Schroder -
>
> >> Maybe the upcoming Hiarcs 6 match we're working on will shed some light.
> >> Maybe we can do this for the top 3-4 programs over the next few months.
> >> It'd be nice to get some real rating info. And maybe we can also get
> >> this folded in to the SSDF list to update it a little with respect to
> >> humans. In any case, it will be interesting to see what happens..
>
> IM is not IM. E.g. if I let Hiarcs6 play vs. IM Bernd Kohlweyer (who
> is living and behaving like Tal, but he has a chess-playing-style like
> Petrosian) or I let Hiarcs6 play against IM Markus Schaefer (who is
> behaving like Petrosian in life and plays like a murderer and Tal'ish
> maniac on the chess-board) the outcome could be totally different.
> But - as I have said - this problem occurs the same way between
> humans. I don't know much about this IM Hergott.
> I know Kohlweyer and Schaefer very good. Does anybody know details
> about playing-style of Hergott?
>
> >Yes.
>
> >- Ed Schroder -
Hergott's style in the openings is like Tony Miles. As for the rest
of the game I am not good enough to comment except that you don't beat
Dean positionally except for his preponderance with pawn sacs. He isn't
an anti-computer player though.
Ingo Althoefer
Chris Whittington wrote:
>Ingo Althoefer <alth...@pdec01.uucp> wrote
>> In November 96 Ed Schroeder had asked a very interesting, hypothetical ques-
>> tion to the SSDF: "Would you allow a program to ask for the identity of its
>> opponents before the games ?" Unfortunately Goran Grottling overreacted,
>> threw Ed Schroeder in one pot with some other guy who got on his nerves,
>> and stopped contributing to our newsgroup. ( What about a return ? )
>> 1.A program with an identification mechanism might perform very well in the
>> SSDF world (and probably also in other niches of our universe): Assume you
>> have two different engines E1 and E2 of same average strength. Against some
>> opponents E1 does better,against others E2. When you have to play a 20 games
>> match against another program A you do the following. In the first half of
>> the match each of your engines plays half of the games, so 5 times E1-A and
>> 5 times E2-A.After this you control the intermediate balances ( this must be
>> the right expression according to my dictionary; what I mean is "the
>> scores" ).
>> If E1 made more points, you use E1 in the other 10 games. If E2 made more
>> points, you let E2 play the other 10 games. ( Of course you are free to
>> compose minor or major variants of this basic scheme. ) This computer E with
>> two engines E1 and E2 and a simple adder should perform better then each Ei
>> alone. ( E1 and E2 may differ only by a few parameters in the evaluation
>> functions. )
>> 2. If you were allowed to make more complicated constructions, the following
>> would yield an SSDF performance of about 2560: Build a very big program BIG
>> including all "traditional" top programs ( Rebell8, Hiarcs 5.6, M-Chess6.0,
>> Genius 3, ...) as engines. Then also feed the current SSDF-list in this
>> program. Not only the ratings, but also the scores of the single matches
>> between "traditional" programs. Now, if BIG has to play against some program
>> X, look in your data which of the engines in BIG has the best score against
>> X. Then play with this program.
>> Taking into account only games against the top 25 programs in the SSDF list,
>> BIG would need only five engines ( = five top programs ) to get a 2560 per-
>> formance. However, adding more programs to BIG would not yield a much better
>> rating.
>> I do not propose to realize idea 2. But No. 1 is not this bad, or ?
>But do the elo improvements in either of these versions mean anything ?
No and yes. No in the sense that this is not a direct way to beating the human
world champions. Yes in an indirect way: Sophisticated variants of this
adaption schemes ( adaption to the opponents ) are required when you want
to have a f u l l y - a u t o m a t i c chess entity which beats world
champions not in single games but in long matches.
>But knowing your opponent ?
Kasparov is fantastic in adapting to special opponents. I believe that without
this ability his Elo number would be at least 100 points lower. ( Perhaps
Gata Kamsky was amongst all top players of the early 90ies the one with the
leastly developed ability to adapt. )
>Ok, so this approach may earn more elo points on the ssdf list. But then
>the Mr Big, and E1 and E2 enter a tourney, separately, each has to play a
>game against a human, and then another human, and then another and so on.
You are right. The basic approach will not help much in round robbin tourna-
ments, at least, when only one tournament is considered.
>Why should this Mr Big play with any more chess skill than E1 or E2. And
>why should Mr Big be more likely to gain more tourney points ?
In the long run he would: After having played ten games against GM Ramon,
he would know which of his engines were most successful against Ramon and
use them more frequently against Ramon.
>It means that Ingo has found a way (amongst all the other ways already
>found) to increase performance on this list because Ingo knows how the list
>is constructed, and how the SSDF make their tests.
My intention to post the two constructions was twofold:
(a) I would have liked the SSDF crew to answer Ed Schroeders question more
constructively, and I wanted to show that there is some potential behind
his question.
(b) More important, the big goal of our computer chess family should be to
beat the human world champions, and this not only in single games but in
properly set up matches. For this goal we need adaption techniques.
My proposals are only small starting steps, but they may help to show
possible directions.
>...
>So Mr Big performs on the SSDF in relation to how the SSDF order the games :)
>... Chris Whittington
Of course.
During the 96 AEGON tournament there were rumors that IGM Seirawan had offered
a 10,000 US-$ bet for a 24 games match against a commercial chess computer
in speed chess ( 5 minutes per game and player ). For such a match it would
be helpful to have several engines and to switch among them from game to
game, so that Seirawan would have more problems to adapt.
Ingo Althoefer.
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
mclane <mcl...@prima.ruhr.de> wrote in article
<E6EL2...@news.prima.ruhr.de>...
Tea is attitude and style. We english have it, you germans don't. Simple as
that.
Plus the water is different, the cups are different, the milk is different,
the tea-bags are too fancy. Believe me, I've tried teaching many germasn
the art of making a good cup of tea. Its hopeless.
And BTW, I don't eat cheese, I eat english beef (well scottish, actually)
with horseradish sauce, roast potatoes, gravy, boiled carrots, yorkshire
pudding and mustard.
Usually washed down with french wine grown on english land in the former
colony of france.
Thorsten, you have about as good a chance to make a good cup of english
tea, as Herr kohl has in getting us english into his euro version of Gross
Deutschland.
Chris Whittington
Tord Kallqvist Romstad
Ed Schroder (rebc...@xs4all.nl) wrote:
: Rebel does not contain software to recognize the opponent.
: Wish I was able... :))
In a match against a computer opponent, this should be easy. When a few
games have been played, examine the openings played. It shouldn't be
difficult to recognize the opening library of the opponent. If your opponent
plays the Bishop opening every second game, you are playing MChess. If it
likes the 1.e4 e5 2.Nf3 Nc6 3.Bb5 f5 gambit, it is probably Genius 5.
Tord
: [ snip ]
: - Ed Schroder -
mclane
"Chris Whittington" <chr...@demon.co.uk> wrote:
>>(Hello Peter Schreiner/Gillgasch
>> and and and ), and I would like to come into the same bloomy mood, but
>> getting a cup of tea, maybe the british way.... like chris always
>> explains me how to do it right, is not easy.
>Tea is attitude and style. We english have it, you germans don't. Simple as
>that.
Grrrrrrrrrrrrrrrr !!! :-(
>Plus the water is different, the cups are different, the milk is different,
>the tea-bags are too fancy. Believe me, I've tried teaching many germasn
>the art of making a good cup of tea. Its hopeless.
I don't use tea-bags. I put the tea in the can where I put hot
(cooking-water) into and wait. I use milk from cows with >= 3.1 % fat.
We tought many britains how to make coffee !! --- sorry - could not
resist ! :-)
>And BTW, I don't eat cheese, I eat english beef (well scottish, actually)
>with horseradish sauce, roast potatoes, gravy, boiled carrots, yorkshire
>pudding and mustard.
I will tast it, if you invite me. I want to find out how it tastes.
>Usually washed down with french wine grown on english land in the former
>colony of france.
Wine ? If beeing drunken makes me happier, i will taste this too. Some
day spock has to be human-like.
>Thorsten, you have about as good a chance to make a good cup of english
>tea, as Herr kohl has in getting us english into his euro version of Gross
>Deutschland.
Hm . No politics. Off - topic !! HAHA ! :-)
>Chris Whittington
Francesco Di Tolla
Robert Hyatt wrote:
> And I still get a "did it again" from Roman
> every few days. Cooking the book isn't enough. The program has got to
> understand those positions (Stonewall is but one well-known example of
> course.) It's a battle of wits, with me on one side, a bunch of IMs and GMs
> on the other, and Crafty right square in the middle.
Hi Bob,
do you keep track of all this games? I'd love to see them. I think it
would
be instructive to see how this can happen, and also how GM play with
computers.
bye
Franz
--
Francesco Di Tolla, Center for Atomic-scale Materials Physics
Physics Departement, Build. 307, Technical Univesity of Denmark,
DK-2800 Lyngby, Denmark, Tel.: (+45) 4525 3208 Fax: (+45) 4593 2399
mailto:dit...@fysik.dtu.dk http://www.fysik.dtu.dk/persons/ditolla.html
Robert Hyatt
Francesco Di Tolla (dit...@fysik.dtu.dk) wrote:
: Robert Hyatt wrote:
: > And I still get a "did it again" from Roman
: > every few days. Cooking the book isn't enough. The program has got to
: > understand those positions (Stonewall is but one well-known example of
: > course.) It's a battle of wits, with me on one side, a bunch of IMs and GMs
: > on the other, and Crafty right square in the middle.
: Hi Bob,
: do you keep track of all this games? I'd love to see them. I think it
: would
: be instructive to see how this can happen, and also how GM play with
: computers.
: bye
: Franz
Actually, I have the PGN for all the games that "my" crafty has ever played
on ICC, FICS and chess.net. hmmm... I may have lost some, now that I
think about it, because some are not in pgn (wrong mailformat). If you'd like
to see a couple posted, let me know. It's easy to do...