| |
Main
Date: 03 Nov 2006 02:40:00
From: RamoNetNet
Subject: Fritz 8 ELO level
|
Can someone tell me the ELO level Fritz 8 whould have - on an "average"
computer ?
Lets say 3 GHz CPU, 1 GB RAM.
Thanks !
|
|
| |
Date:
From: Martin Brown
Subject: Re: Fritz 8 ELO level
|
|
| |
Date: 04 Nov 2006 17:10:39
From:
Subject: Re: Fritz 8 ELO level
|
RamoNetNet wrote:
> Can someone tell me the ELO level Fritz 8 whould have - on an "average"
> computer ?
> Lets say 3 GHz CPU, 1 GB RAM.
> Thanks !
Check out http://web.telia.com/~u85924109/ssdf/list.htm
it's a compilation of programs and ratings for various computer chess
programs. The ssdf's ratings don't coincide with elo ratings for human
play, but they give you some objective metric for measuring strength
between individual programs.
|
| | |
Date: 06 Nov 2006 11:28:24
From: David Richerby
Subject: Re: Fritz 8 ELO level
|
<[email protected] > wrote:
> RamoNetNet wrote:
>> Can someone tell me the ELO level Fritz 8 whould have - on an "average"
>> computer ?
>
> Check out http://web.telia.com/~u85924109/ssdf/list.htm
>
> it's a compilation of programs and ratings for various computer chess
> programs. The ssdf's ratings don't coincide with elo ratings for human
> play, but they give you some objective metric for measuring strength
> between individual programs.
Yes, it's important to realize that being rated, say, 2800 on that
list doesn't mean ``is as good as Topalov'' but ``is as likely to beat
a program rated 2600 on that list as Topalov is to beat a human rated
2600 FIDE.''
Dave.
--
David Richerby Aquatic Perforated Gerbil (TM): it's
www.chiark.greenend.org.uk/~davidr/ like a children's pet but it's full
of holes and it lives in the sea!
|
| | | |
Date: 07 Nov 2006 10:37:41
From: Ange1o DePa1ma
Subject: Re: Fritz 8 ELO level
|
"David Richerby" <[email protected] > wrote
>> RamoNetNet wrote:
>>> Can someone tell me the ELO level Fritz 8 whould have - on an "average"
>>> computer ?
>>
>> Check out http://web.telia.com/~u85924109/ssdf/list.htm
>>
>> it's a compilation of programs and ratings for various computer chess
>> programs. The ssdf's ratings don't coincide with elo ratings for human
>> play, but they give you some objective metric for measuring strength
>> between individual programs.
>
> Yes, it's important to realize that being rated, say, 2800 on that
> list doesn't mean ``is as good as Topalov'' but ``is as likely to beat
> a program rated 2600 on that list as Topalov is to beat a human rated
> 2600 FIDE.''
Seems like we can't avoid philosophy in this newsgroup.
Computer ratings do appear to be generated from a different "player" pool
than human ratings. Then there is the horizon effect, which occasionally
allows even someone as low-rated as myself to out-calculate a computer.
That being said, there is no reason to think that a computer rated the same
as Topalov would not be every bit as strong. Otherwise, it would have a
lower rating -- all things being equal.
We've had this "human vs. computer" discussion before. I know I am in the
minority. The last two big human-computer matches ended in draws. I don't
know how the machines' ratings changed after those matches, but it appears
they are every bit as strong as the world's best human players.
If there was a way to beat those two computers I am certain that Kramnik and
Kasparov would have done so. Their incentives were huge.
|
| | | | |
Date: 07 Nov 2006 17:33:29
From: David Richerby
Subject: Re: Fritz 8 ELO level
|
Ange1o DePa1ma <[email protected] > wrote:
> "David Richerby" <[email protected]> wrote
>> Yes, it's important to realize that being rated, say, 2800 on [the
>> SSDF rating list] list doesn't mean ``is as good as Topalov'' but
>> ``is as likely to beat a program rated 2600 on that list as Topalov
>> is to beat a human rated 2600 FIDE.''
>
> Seems like we can't avoid philosophy in this newsgroup.
This isn't philosophy. It's a discussion of what the rating systems
(deliberate plural!) do and do not say.
> Computer ratings do appear to be generated from a different "player"
> pool than human ratings.
They appear to be generated from a different player pool because they
*are* generated from a different player pool. The SSDF rating list is
generated entirely from computer-computer games; the FIDE rating list
is generated entirely from human-human games.
> That being said, there is no reason to think that a computer rated
> the same as Topalov would not be every bit as strong.
Only if it got that rating by playing the same opponents as Topalov.
Which it didn't.
> Otherwise, it would have a lower rating -- all things being equal.
No! How many times do I have to say it? Rating. Measures.
Performance. If I beat ten people rated 100 points below me, my
rating changes by exactly the same amount as anyone else's would if
they beat ten people rated 100 points below them.
> We've had this "human vs. computer" discussion before. I know I am
> in the minority. The last two big human-computer matches ended in
> draws. I don't know how the machines' ratings changed after those
> matches, but it appears they are every bit as strong as the world's
> best human players.
Yes, the top computers are about as strong as the top humans. I'm not
disputing that conclusion. But you cannot legitimately reach this
conclusion by looking at the SSDF and FIDE rating lists and saying
that Topalov is rated 2813 and Fritz 9 is rated 2811 so the two are
equally strong.
Quite apart from anything else, FIDE could add 1000 points to
everybody's rating without changing the meaning of the rating system
one iota. The only significance in rating calculations is the
*difference* between ratings: it makes not one jot of difference
whether Topalov is rated 2813, 3813 or 3000813, or even a nice round
10000000, as long as Anand is 34 points behind, Kramnik and Svidler
another 29 points back and so on. The numbers are completely
arbitrary.
Dave.
--
David Richerby Unholy Goldfish (TM): it's like a fish
www.chiark.greenend.org.uk/~davidr/ but it's also a crime against nature!
|
| | | | | |
Date: 07 Nov 2006 14:51:37
From: Kenneth Sloan
Subject: Re: Fritz 8 ELO level
|
David Richerby wrote:
>
> No! How many times do I have to say it? Rating. Measures.
> Performance. If I beat ten people rated 100 points below me, my
> rating changes by exactly the same amount as anyone else's would if
> they beat ten people rated 100 points below them.
>
Not necessarily. Depends on 'k'.
--
Kenneth Sloan [email protected]
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/
|
| | | | | | |
Date: 13 Nov 2006 11:54:46
From: David Richerby
Subject: Re: Fritz 8 ELO level
|
Kenneth Sloan <[email protected] > wrote:
> David Richerby wrote:
>> No! How many times do I have to say it? Rating. Measures.
>> Performance. If I beat ten people rated 100 points below me, my
>> rating changes by exactly the same amount as anyone else's would if
>> they beat ten people rated 100 points below them.
>
> Not necessarily. Depends on 'k'.
Good point. Add an ``assuming the same k-value'' to my last sentence.
Dave.
--
David Richerby Homicidal Priest (TM): it's like a man
www.chiark.greenend.org.uk/~davidr/ of the cloth but it wants to kill you!
|
| | | | | |
Date: 07 Nov 2006 22:45:14
From: Ange1o DePa1ma
Subject: Re: Fritz 8 ELO level
|
Maybe I read too much into the original post. I assumed it was heading into
"who is stronger" territory, with a twist, eg "for a given ELO who is
stronger?"
Sorry, I read too quickly. Here's the post I was responding to:
<[email protected] > wrote:
Yes, it's important to realize that being rated, say, 2800 on that
list doesn't mean ``is as good as Topalov'' but ``is as likely to beat
a program rated 2600 on that list as Topalov is to beat a human rated
2600 FIDE.''
What he's saying is that a 200 point difference for computers is equivalent
to a 200 point difference for humans since they both use Elo's formulas.
No-one could disagree with this. And of course since the ratings were
generated differently they are not directly, precisely comparable although
differences between ratings are.
|
| | | | |
Date: 07 Nov 2006 10:08:16
From: Kenneth Sloan
Subject: Re: Fritz 8 ELO level
|
Ange1o DePa1ma wrote:
>
> Seems like we can't avoid philosophy in this newsgroup.
>
What's wrong with philosophy?
> Computer ratings do appear to be generated from a different "player" pool
> than human ratings. Then there is the horizon effect, which occasionally
> allows even someone as low-rated as myself to out-calculate a computer.
The horizon effect is completely irrelevant here. Are you just tossing
in buzzwords, hoping one might fit?
>
> That being said, there is no reason to think that a computer rated the same
> as Topalov would not be every bit as strong. Otherwise, it would have a
> lower rating -- all things being equal.
One of the things that would have to be equal would be the rating
system. The computer ratings under discussion are generated by a
different organization, using different methods, and rating different
games. NOTHING is "equal".
>
> We've had this "human vs. computer" discussion before. I know I am in the
> minority. The last two big human-computer matches ended in draws. I don't
> know how the machines' ratings changed after those matches, but it appears
> they are every bit as strong as the world's best human players.
That wasn't the question.
>
> If there was a way to beat those two computers I am certain that Kramnik and
> Kasparov would have done so. Their incentives were huge.
>
>
What does that have to do with differences between rating systems???
--
Kenneth Sloan [email protected]
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/
|
| | | |
Date: 06 Nov 2006 15:36:58
From: Ruud
Subject: Re: Fritz 8 ELO level
|
"David Richerby" <[email protected] > schreef in bericht
news:mqn*[email protected]...
> <[email protected]> wrote:
>> RamoNetNet wrote:
>>> Can someone tell me the ELO level Fritz 8 whould have - on an "average"
>>> computer ?
>>
>> Check out http://web.telia.com/~u85924109/ssdf/list.htm
>>
>> it's a compilation of programs and ratings for various computer chess
>> programs. The ssdf's ratings don't coincide with elo ratings for human
>> play, but they give you some objective metric for measuring strength
>> between individual programs.
>
> Yes, it's important to realize that being rated, say, 2800 on that
> list doesn't mean ``is as good as Topalov'' but ``is as likely to beat
> a program rated 2600 on that list as Topalov is to beat a human rated
> 2600 FIDE.''
Can I just say that i have my doubts about this theory? ELO is, to my
knowledge, for the purpose of determining the playing strength in chess.
Of course, there aren't many games available to compare chess engines'
strenght to human playing strength (and why should there ever be?).
But why compare ELO's to begin with? On several chess servers, there are
humans, as well as engines, reaching ELO of 3200.
All I'm saying is, that it strikes me as somewhat unfare, that before
engines were anywhere near human GM's, everyone was happy to compare their
elo's, and now they're better, we should all let go of comparing them.
Maybe it's better to let go of the elo-system to measure playing strenght
alltogether.
|
| | | | |
Date: 07 Nov 2006 10:10:32
From: David Richerby
Subject: Re: Fritz 8 ELO level
|
Ruud <[email protected] > wrote:
> "David Richerby" <[email protected]> schreef:
>> Yes, it's important to realize that being rated, say, 2800 on [the
>> SSDF list] doesn't mean ``is as good as Topalov'' but ``is as
>> likely to beat a program rated 2600 on that list as Topalov is to
>> beat a human rated 2600 FIDE.''
>
> Can I just say that i have my doubts about this theory?
You can say whatever you like! But in this case, you'd be wrong. :-)
> ELO is, to my knowledge, for the purpose of determining the playing
> strength in chess.
No. Elo measures the performance (proportion of wins, draws, losses)
of players within a given pool. They could be playing chess or tennis
or flipping beer mats -- it really doesn't matter. All an Elo rating
allows you to do is to say how likely one player is to beat another in
the pool.
Here's another example. FIDE and the USCF both use Elo's system to
rate chess players. There are many players who have both a FIDE
rating and a USCF rating but these numbers are usually not the same --
as I recall, it's not at all unusual for them to differ by around 100
points or so. For example, here are the top ten in the USCF October
2006 rating list:
Name USCF FIDE Diff
---------------------------------------
Kamsky, Gata 2753 2705 48
Nakamura, Hikaru 2709 2640 69
Onischuk, Alexander 2702 2668 34
Kaidanov, Gregory 2692 2595 97
Ibragimov, Idlar 2682 2616 68
Shulman, Yury 2679 2614 65
Ehlvest, Jaan 2666 2597 69
Christiansen, Larry 2663 2572 91
Gulko, Boris 2661 2570 91
Akobian, Varuzhan 2654 2598 56
If Elo measured chess strength, you'd expect those numbers to be much
closer, I think. Now, within any one pool, playing strength should be
fairly well correlated with results but you can't go any further than
that.
> Of course, there aren't many games available to compare chess
> engines' strenght to human playing strength (and why should there
> ever be?). But why compare ELO's to begin with?
Because the original poster wanted an estimate of the strength of
chess programs compared to the top human players. Elo ratings are the
only real approximation to chess strength that we have. Indeed, an
Elo rating pool including the top GMs and the top computers would give
a reasonable indication of their relative strengths. But we don't
have such a rating pool.
> All I'm saying is, that it strikes me as somewhat unfare, that
> before engines were anywhere near human GM's, everyone was happy to
> compare their elo's, and now they're better, we should all let go of
> comparing them.
I'm sorry but I don't understand what you're saying here. Regardless
of the relative strength of humans and computers, it would be wrong to
compare SSDF Elo ratings with human chess federation ratings because
they're not measuring the same thing. But, regardless of the relative
strength of humans and computers, people still want to know how those
strengths compare. It doesn't make any difference that, instead of
asking the question ``How far down the FIDE list do we have to go
before we find a GM who can't beat the best computer?'', we'll soon be
asking ``How far down the SSDF list do we have to go before we find a
computer that can't beat the best GM?''
> Maybe it's better to let go of the elo-system to measure playing
> strenght alltogether.
Nobody has anything better than ratings based on performance.
Dave.
--
David Richerby Hungry Tool (TM): it's like a hammer
www.chiark.greenend.org.uk/~davidr/ but it'll eat you!
|
| | | | | |
Date: 07 Nov 2006 12:44:12
From: David Kane
Subject: Re: Fritz 8 ELO level
|
"David Richerby" <[email protected] > wrote in message
news:4yi*[email protected]...
>
> Because the original poster wanted an estimate of the strength of
> chess programs compared to the top human players. Elo ratings are the
> only real approximation to chess strength that we have. Indeed, an
> Elo rating pool including the top GMs and the top computers would give
> a reasonable indication of their relative strengths. But we don't
> have such a rating pool.
This is not clear. There have been games between humans and computers.
The fact that, for administrative reasons, the SSDF and FIDE don't rate
each other's games, or computer-human games, doesn't mean that
the (larger) pool can't be rated.
If your point is simply that they *haven't* done so, then you are correct.
But I don't think that was the question.
|
| | | | |
Date: 07 Nov 2006 01:17:38
From: Ralf Callenberg
Subject: Re: Fritz 8 ELO level
|
06.11.2006 15:36, Ruud:
>> Yes, it's important to realize that being rated, say, 2800 on that
>> list doesn't mean ``is as good as Topalov'' but ``is as likely to beat
>> a program rated 2600 on that list as Topalov is to beat a human rated
>> 2600 FIDE.''
>
> Can I just say that i have my doubts about this theory?
That's not a theory, that is what ELO is about. It only measures
relative strength. Let's say you go to a big elementary school and teach
the kids chess - and then let them play rated games. At the beginning,
everybody gets, let's say 1800. Some kids will be better than others,
and after a while their numbers will be distributed, more or less like a
Gauss curve. The majority will stay around 1800, but at the top their
will be some guys with around 2600 or 2800. That happens automatically,
without any geniuses around.
The numbers for the programs are not as uncorrelated to those from FIDE,
because some programs already played games against GMs, so you can
estimate, that the best programs indeed have a FIDE rating somewhere
above 2600, this provides some kine of "gauging".
Greetings,
Ralf
|
| | | | | |
Date: 06 Nov 2006 20:53:23
From: Kenneth Sloan
Subject: Re: Fritz 8 ELO level
|
Ralf Callenberg wrote:
> 06.11.2006 15:36, Ruud:
>
>>> Yes, it's important to realize that being rated, say, 2800 on that
>>> list doesn't mean ``is as good as Topalov'' but ``is as likely to beat
>>> a program rated 2600 on that list as Topalov is to beat a human rated
>>> 2600 FIDE.''
>>
>> Can I just say that i have my doubts about this theory?
>
> That's not a theory, that is what ELO is about. It only measures
> relative strength. Let's say you go to a big elementary school and teach
> the kids chess - and then let them play rated games. At the beginning,
> everybody gets, let's say 1800. Some kids will be better than others,
> and after a while their numbers will be distributed, more or less like a
> Gauss curve.
Your first error. There's no especially good reason to expect the
population of players to generate a "Gauss curve". Elo assumes that
a player's individual performances will generate a Gaussian distribution
- but the ratings for a large population of players may, or may not, be
well described by a Gaussian. The USCF population, for example, is at
least bi-modal, and may well be tri-modal.
> The majority will stay around 1800,
Not neccessarily so.
> but at the top their
> will be some guys with around 2600 or 2800. That happens automatically,
> without any geniuses around.
maybe. I suspect that 800-1000 Elo points is an awful large spread for
a pool as small as a single Elementary school.
but, it's possible. So are ratings of 10,000. and negative 5000.
--
Kenneth Sloan [email protected]
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/
|
| | | | | | |
Date: 07 Nov 2006 09:38:54
From: David Richerby
Subject: Minimum ratings (was Re: Fritz 8 ELO level)
|
[ This has nothing to do with computer chess any more so let's shift
it over to rgc.misc. ]
Kenneth Sloan <[email protected] > wrote:
> but, it's possible. So are ratings of 10,000. and negative 5000.
In a pure Elo system, yes. But doesn't the USCF have a minimum rating
of zero (or was it 100)? IIRC, the BCF doesn't let people have
negative ratings, either, but they don't use Elo. The issue doesn't
come up with FIDE because they only maintain ratings over 1800, IIRC.
Dave.
--
David Richerby Sumerian Painting (TM): it's like
www.chiark.greenend.org.uk/~davidr/ a Renaissance masterpiece that's
really old!
|
| | | | |
Date: 06 Nov 2006 09:22:21
From: Kenneth Sloan
Subject: Re: Fritz 8 ELO level
|
Ruud wrote:
>
> Can I just say that i have my doubts about this theory? ELO is, to my
> knowledge, for the purpose of determining the playing strength in chess.
No, it's not. It's for the purpose of measuring past performance and
predicting future performance.
Performances are determined by many factors. "Chess strength" (whatever
that is) is an important factor - but not the only one.
--
Kenneth Sloan [email protected]
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/
|
| |
Date: 04 Nov 2006 10:30:30
From: Gilles Garrigues
Subject: Re: Fritz 8 ELO level
|
about 2650 ?
"RamoNetNet" <[email protected] > a �crit dans le message de news:
[email protected]...
>
> Can someone tell me the ELO level Fritz 8 whould have - on an "average"
> computer ?
> Lets say 3 GHz CPU, 1 GB RAM.
> Thanks !
>
|
|
