The Crafty-Rebel NPS challenge started!!!
Ed Schroder
The Crafty-Rebel NPS challenge started!!!
Today the Cdrom with the Crafty big book (90 Mb) finally arrived (thanks
Moritz!) and the 1:100 match has started.
The comparison 50-60 game match on equal tournament time (4:30 average)
will start in a few days too.
You will find the match at:
http://www.xs4all.nl/~rebchess/match.htm
Make your bets now!
- Ed Schroder -
Peter W. Gillgasch
Ed Schroder <rebc...@xs4all.nl> wrote:
> Make your bets now!
Crafty will surely kill Rebel at this time odds match.
I am betting one bottle of Bavarian beer. Anybody ?
-- Peter
May God grant me the serenity to accept the things I cannot change,
courage to choke the living shit out of those who piss me off,
and wisdom to know where I should hide the bodies...
Robert Hyatt
Peter W. Gillgasch (gil...@ilk.de) wrote:
: Ed Schroder <rebc...@xs4all.nl> wrote:
: > Make your bets now!
: Crafty will surely kill Rebel at this time odds match.
: I am betting one bottle of Bavarian beer. Anybody ?
Nope. If Crafty loses, I think something is badly wrong in my eval,
but I'll wait to see. However, draws are likely. A lot's going to depend
on the positions the book leaves us in. If it's like the CM5000 game in
the "Kup", I'll be more than happy. :) But it is likely that at least
a game or two or three will leave the book in an endgame... The question
will be is it the type of endgame that is simply drawn, period.
I haven't looked at Ed's home page yet, but hope he can give the eval
and depth for the moves...
brucemo
Robert Hyatt wrote:
>
> Peter W. Gillgasch (gil...@ilk.de) wrote:
> : Ed Schroder <rebc...@xs4all.nl> wrote:
>
> : > Make your bets now!
>
> : Crafty will surely kill Rebel at this time odds match.
>
> : I am betting one bottle of Bavarian beer. Anybody ?
>
> Nope. If Crafty loses, I think something is badly wrong in my eval,
> but I'll wait to see. However, draws are likely. A lot's going to depend
> on the positions the book leaves us in. If it's like the CM5000 game in
> the "Kup", I'll be more than happy. :) But it is likely that at least
> a game or two or three will leave the book in an endgame... The question
> will be is it the type of endgame that is simply drawn, period.
You might lose some games. But draws are certainly very likely.
I did an 8-game match with your Crafty on chess.net (p6/200) versus Spinach, which
is Ferret on a P5/133, thinking for 50 milliseconds, no pondering. Got blanked.
I then did a 10-game five-minute blitz match Crafty vs Pinky, which is the same
thing thinking for 167 milliseconds, and scored half a point, as both of them came
out of book with a negative score, and they happened to decide to do a 3x
repetition right away.
Mike Byrne did a match with a full-out Crafty versus Pinky and Pinky managed to get
like 20%, but Byrne was not using a P6.
My intent was to show that sometimes a good program will defeat another good
program even when it is at a severe speed disadvantage. I certainly didn't show
that in this recent experiment, but I think I may try again.
I certainly saw a lot of good positions go up in smoke due to massive tactics.
bruce
Robert Hyatt
brucemo (bru...@nwlink.com) wrote:
Those are the ones that doubling *both* sides time will tend to clear up. If you
are doing 100ms searches, tactics are nasty. if you were playing 5 mins per move
vs the equivalent extension for crafty, my tests generally find that the # of
draws goes up a good bit, and the weaker side will occasionally toss in a win too.
The thing Ed and I will find out is not just a small difference, but what happens
at deep depths, if one side gets an additional 3-4 plies. That's not been tested
anywhere that I know of. In fact, once we get the data, we ought to write this up
for the JICCA, as another chapter in the tests that have been done in the past...
Tom C. Kerrigan
> > Make your bets now!
> Crafty will surely kill Rebel at this time odds match.
> I am betting one bottle of Bavarian beer. Anybody ?
> -- Peter
Eew. I'll put a bottle of Koelsch on Crafty.
Cheers,
Tom
dong...@aol.com
In article <3322F2...@nwlink.com>, brucemo <bru...@nwlink.com> writes:
>Robert Hyatt wrote:
>>
>> Peter W. Gillgasch (gil...@ilk.de) wrote:
>> : Ed Schroder <rebc...@xs4all.nl> wrote:
>>
>> : > Make your bets now!
>>
>> : Crafty will surely kill Rebel at this time odds match.
>>
>> : I am betting one bottle of Bavarian beer. Anybody ?
>>
>bruce
>
>
Interesting. I have recently played around with time handicap games myself and have discovered that perhaps "raw speed" is the current "end all be all" performance factor in chess computers today.
Your findings seem to support this as well?
In the test I ran, I went a step further by pitting a relic (RadioShack ChessChampion 2150 cir.1988) against a modern world computer champ, MChess Pro 5.0 on my P75.
I determined the handicap of :02 seconds for MCP5 to be the most equal with CC2150's :30 seconds per move level, in an attempt to nullify "raw speed" as a performance factor.
After 10 games, the score sits at 5-5!
Conclusions?
1.)In chess, all things being equal, there is strong parity amongst programs of today and yesterday, but the Hyatt axiom "Speed Kills" is still an absolute. :-(
2.)Ed, you are an eternal optimist. ;-)
3.)I won't be the guy buying anyone a bavarian beer. Rebel 8 may not draw a single game, let a lone win one.
yours in chess,
Don
Ramsey MN USA
brucemo
dong...@aol.com wrote:
> 3.)I won't be the guy buying anyone a bavarian beer. Rebel 8 may not draw a single game, let a lone win one.
Ah. I would never go this far though. The games I was playing were blitz games, so
at 1/6 second per move my program was getting almost no depth.
You may get different results in super-blitz vs blitz than you would at blitz vs
standard, or standard vs correspondence, because you get enough depth to play
reasonable chess.
If I'm playing super-blitz vs blitz, the positions are full of shallow tactics that
I won't see. There will still be deep tactics at standard vs correspondence, but
there may be fewer of them.
I can think of a couple more interesting experiments, one of which Ed and Bob are
already doing. I'll be interested to see if Rebel can get anything against Crafty
in any of these games. It may find tactics that Crafty doesn't see, but I think it
is more likely that it may reach strategically safe situations and draw, or reach
positions where it has an advantage, and push through for a very difficult win. I
don't know how Rebel and Crafty compare with each other, but I think it would be
likely that Crafty could do the same thing to Rebel were the situation reversed.
I'd also be interested in seeing what happens to the draw rate when two programs
play games at various speeds (not handicap games, but instead games where both
programs are using the same time control). For instance, if two programs play
super-blitz against each other, would draws be less likely than if they play
standard against each other? More draws may indicate a reduced chance of finding
knock-out tactics. In autoplay 5 0 games I have noticed a lot of draws, but in
autoplay "game in 15 second" games, draws seem to be rare.
This is all related to the notion of "tactical sufficiency", as recently debated at
length in this newsgroup. I think that it is likely that there is such a thing, but
that rather than being at some defined point, it may manifest as a reduced liklihood
of knockout tactics existing as search depth increases. I haven't done any
experiments though, I just have some ideas.
Another experiment might involve studying the liklihood that a program will change
its PV move at a particular depth. I bet changes between depth 1 and depth 2 are
very common, but I doubt you see as many between depth 8 and depth 9. If you never
change your PV move you aren't playing any more strongly if you think longer, so
perhaps this is some of the source of this notion that computers don't get much
stronger if they search a lot deeper (longer).
I would like to see examples where programs find POSITIONAL moves after having done
deep searches. For instance, a move is considered to be positionally "right", but a
program doesn't plan to play it until it's working on a 10-ply search or soemthing,
then suddenly it switches to that move for the first time. I think this might be
indicative of a program really understanding the positional elements present. I
think I would trust this more than I would seeing a program choose some positionally
"right" move at depth 2 and sticking with it permanently.
An example might be BK position #2, in which some programs play d5 immediately, but
I've never heard of any that would switch to it after thinking for five minutes.
3r1k2/4npp1/1ppr3p/p6P/P2PPPP1/1NR5/5K2/2R5 w - - 0 1
A program that would play d5 quickly here is demonstrating "instinct", not
"understanding". I wonder if instinct is really worth very much?
I think there may have been some "findable" deep positional problems in the CCR test
suite developed by Kaufmann.
Once again, I have no experimental data, just ideas.
bruce
Robert Hyatt
Tom C. Kerrigan (kerr...@merlin.pn.org) wrote:
: Peter W. Gillgasch (gil...@ilk.de) wrote:
: > Ed Schroder <rebc...@xs4all.nl> wrote:
: > > Make your bets now!
: > Crafty will surely kill Rebel at this time odds match.
: > I am betting one bottle of Bavarian beer. Anybody ?
: > -- Peter
: Eew. I'll put a bottle of Koelsch on Crafty.
: Cheers,
: Tom
I'm going to regret this, but what the hell is a bottle of koelsch?
:)
Robert Hyatt
brucemo (bru...@nwlink.com) wrote:
: dong...@aol.com wrote:
: > 3.)I won't be the guy buying anyone a bavarian beer. Rebel 8 may not draw a single game, let a lone win one.
: Ah. I would never go this far though. The games I was playing were blitz games, so
: at 1/6 second per move my program was getting almost no depth.
: You may get different results in super-blitz vs blitz than you would at blitz vs
: standard, or standard vs correspondence, because you get enough depth to play
: reasonable chess.
: If I'm playing super-blitz vs blitz, the positions are full of shallow tactics that
: I won't see. There will still be deep tactics at standard vs correspondence, but
: there may be fewer of them.
: I can think of a couple more interesting experiments, one of which Ed and Bob are
: already doing. I'll be interested to see if Rebel can get anything against Crafty
: in any of these games. It may find tactics that Crafty doesn't see, but I think it
: is more likely that it may reach strategically safe situations and draw, or reach
: positions where it has an advantage, and push through for a very difficult win. I
: don't know how Rebel and Crafty compare with each other, but I think it would be
: likely that Crafty could do the same thing to Rebel were the situation reversed.
This is the one "weak" link in our experiment. If Ed has some neat eval
trick that I don't know about in crafty, then he might play into a position
that crafty thinks is dead even, while Rebel realizes that it is dead won.
That can always happen. Might happen the other way too of course, but
that wouldn't be a surprise since crafty already has a 3+ ply search
advantage...
: I'd also be interested in seeing what happens to the draw rate when two programs
: play games at various speeds (not handicap games, but instead games where both
: programs are using the same time control). For instance, if two programs play
: super-blitz against each other, would draws be less likely than if they play
: standard against each other? More draws may indicate a reduced chance of finding
: knock-out tactics. In autoplay 5 0 games I have noticed a lot of draws, but in
: autoplay "game in 15 second" games, draws seem to be rare.
: This is all related to the notion of "tactical sufficiency", as recently debated at
: length in this newsgroup. I think that it is likely that there is such a thing, but
: that rather than being at some defined point, it may manifest as a reduced liklihood
: of knockout tactics existing as search depth increases. I haven't done any
: experiments though, I just have some ideas.
: Another experiment might involve studying the liklihood that a program will change
: its PV move at a particular depth. I bet changes between depth 1 and depth 2 are
: very common, but I doubt you see as many between depth 8 and depth 9. If you never
: change your PV move you aren't playing any more strongly if you think longer, so
: perhaps this is some of the source of this notion that computers don't get much
: stronger if they search a lot deeper (longer).
: I would like to see examples where programs find POSITIONAL moves after having done
: deep searches. For instance, a move is considered to be positionally "right", but a
: program doesn't plan to play it until it's working on a 10-ply search or soemthing,
: then suddenly it switches to that move for the first time. I think this might be
: indicative of a program really understanding the positional elements present. I
: think I would trust this more than I would seeing a program choose some positionally
: "right" move at depth 2 and sticking with it permanently.
Good idea. Here is *your* mission. Find a neat positional game. Not one
with a lot of tactical fireworks. IE, find a karpov vs someone game. Let's
let a group of programs "annotate" that game with *very* deep searches. And
then lets produce some stats for each program, to answer two questions:
1. as time extends, what's the chances of finding a different move;
2. ditto, but find a *better* move. :)
would be *very* interesting to see if 4x the time finds a better move.
However, we'll need to post the annotated game and get some good chess
players to look over them as well to see if the moves are better or worse.
Perfect place for Martin you-know-who here. :)
: An example might be BK position #2, in which some programs play d5 immediately, but
: I've never heard of any that would switch to it after thinking for five minutes.
: 3r1k2/4npp1/1ppr3p/p6P/P2PPPP1/1NR5/5K2/2R5 w - - 0 1
: A program that would play d5 quickly here is demonstrating "instinct", not
: "understanding". I wonder if instinct is really worth very much?
and some play it because it is known to be the right move (supposedly) and
they've been tuned to do so. :)
: I think there may have been some "findable" deep positional problems in the CCR test
Robert Hyatt
dong...@aol.com wrote:
: In article <3322F2...@nwlink.com>, brucemo <bru...@nwlink.com> writes:
: >Robert Hyatt wrote:
: >>
: >> Peter W. Gillgasch (gil...@ilk.de) wrote:
: >> : Ed Schroder <rebc...@xs4all.nl> wrote:
: >>
: >> : > Make your bets now!
: >>
: >> : Crafty will surely kill Rebel at this time odds match.
: >>
: >> : I am betting one bottle of Bavarian beer. Anybody ?
: >>
: >> Nope. If Crafty loses, I think something is badly wrong in my eval,
: Conclusions?
: 1.)In chess, all things being equal, there is strong parity amongst programs of today and yesterday, but the Hyatt axiom "Speed Kills" is still an absolute. :-(
: 2.)Ed, you are an eternal optimist. ;-)
: 3.)I won't be the guy buying anyone a bavarian beer. Rebel 8 may not draw a single game, let a lone win one.
: yours in chess,
: Don
I agree about the huge handicap. a very good test is to take crafty,
with ponder=off, and give one copy 2x the time of the other. The
win/loss ratio is usually quite significant...
I don't agree about draws. The book will play an imortant role here,
although in endgames, that 3+ ply search advantage might translate into
6-10... which *might* be important...
Moritz Berger
On 10 Mar 1997 02:02:30 GMT, hy...@crafty.cis.uab.edu (Robert Hyatt)
wrote:< snip >
>: Eew. I'll put a bottle of Koelsch on Crafty.
>: Cheers,
>: Tom
>I'm going to regret this, but what the hell is a bottle of koelsch?
beer like drink in the area of cologne (Koeln). The name means
"something from Koeln", which is a better description for it than beer
;-) Some people say its more a kind of lemonade with alcohol ... It's
usually served in long, thin glasses in a quantity of 0,2l.
Wow, now I will be flamed ... ;-))))
Moritz
-------------
Moritz...@msn.com
Peter W. Gillgasch
Tom C. Kerrigan <kerr...@merlin.pn.org> wrote:
> Eew. I'll put a bottle of Koelsch on Crafty.
At least one guy with a sense for the important things
in life on this newsgroup...
Bring the bottle to the next world micro. Bring some for yourself, you
will need it 8^)
Peter W. Gillgasch
<dong...@aol.com> wrote:
> After 10 games, the score sits at 5-5!
Interesting.
> Conclusions? 1.)In chess, all things being equal, there is strong parity
> amongst programs of today and yesterday, but the Hyatt axiom "Speed Kills"
> is still an absolute. :-(
SPEED SPEED SPEED !!! We have plenty. But still we are not satisfied 8^)
> 2.)Ed, you are an eternal optimist. ;-)
I agree. Nevertheless it is a brave thing to do for Ed. I mean, if it
will be the massacre most of us are expecting it takes a man to post
those games on his commercial web site.
> 3.)I won't be the guy buying anyone a bavarian beer.
Too bad 8^) Well, I'll drink Tom's Koelsch then... More takers ?
> Rebel 8 may not draw a single game, let a lone win one.
That's a bit unlikely. I guess that in the end it will be 8-2 for
Crafty. 1 loss, 2 draws (shit happens even to deep searchers).
Albert Silver
In article <5fvqcm$r...@juniper.cis.uab.edu>,
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
> : In the test I ran, I went a step further by pitting a relic (RadioShack
ChessChampion 2150 cir.1988) against a modern world computer champ,
MChess Pro 5.0 on my P75.
>
> : I determined the handicap of :02 seconds for MCP5 to be the most equal with
CC2150's :30 seconds per move level, in an attempt to nullify "raw speed"
as a performance factor.
>
> : After 10 games, the score sits at 5-5!
>
> : Conclusions?
> : 1.)In chess, all things being equal, there is strong parity amongst programs
of today and yesterday, but the Hyatt axiom "Speed Kills" is still an
absolute. :-(
>
> : 2.)Ed, you are an eternal optimist. ;-)
>
> : 3.)I won't be the guy buying anyone a bavarian beer. Rebel 8 may not draw a
single game, let a lone win one.
>
> : yours in chess,
> : Don
>
> I agree about the huge handicap. a very good test is to take crafty,
> with ponder=off, and give one copy 2x the time of the other. The
> win/loss ratio is usually quite significant...
>
> I don't agree about draws. The book will play an imortant role here,
> although in endgames, that 3+ ply search advantage might translate into
> 6-10... which *might* be important...
Although I'm not arguing that Rebel will win the handicapped match, I
don't think the match giving Mchess 2 seconds is really significant to
disprove Ed. As I recall, Bob told us that when your machine is
calculating 14-16 plys every move, adding another couple of plys isn't
going to give the machine a huge difference in strength. Though when you
go from 4 plys to 6, then it IS significant. That's why Ed only accepted
the wager if Rebel could think for 3-4 minutes at which point it will see
quite a bit, and he doesn't think that just adding a few plys will give a
weaker program (no offence Bob) a crushing edge. Still, though it is
interesting, I don't think that this will prove anything as far as Deep
Blue is concerned because as Bob pointed out, it suffers NO slowdown no
matter how many algorithms it uses, and that is definitely not the case
for programs running on microcomputers. Getting back to the Mchess match:
give IT 2-3 minutes and the other the same proportionate advantage in
time, and see if it still only manages to draw the match (you can even
repeat the openings), and I'll bet heavily in favor of Mchess.
Albert Silver
-------------------==== Posted via Deja News ====-----------------------
http://www.dejanews.com/ Search, Read, Post to Usenet
Albert Silver
In article <5fvqoc$r...@juniper.cis.uab.edu>,
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
> Good idea. Here is *your* mission. Find a neat positional game. Not one
> with a lot of tactical fireworks. IE, find a karpov vs someone game.
> : bruce
How about Fischer - Reschevsky US Championship 1962-3 (game no.43 in his
60 memorable games)? That game always struck me as being an absolute
masterpiece in position play and clarity.
Robert Hyatt
Peter W. Gillgasch (gil...@ilk.de) wrote:
: That's a bit unlikely. I guess that in the end it will be 8-2 for
: Crafty. 1 loss, 2 draws (shit happens even to deep searchers).
My take is slightly different. I'm expecting something like (just
a wild guess here) 6 wins, 4 draws... or 6 wins, 3 draws, 1 loss,
although the loss is just to be safe. If you throw out the draws,
and only count wins, it might be closer to your 9-1 result, and
the result I'd expect against DB. However, there are some positions
where depth doesn't matter... only a really stupid player could lose
some hopeless draws (some Petrov lines come to mind where the queens
and most pieces are gone with equal pawn structure remaining.
Peter W. Gillgasch
Tom C. Kerrigan <kerr...@merlin.pn.org> wrote:
> Eew. I'll put a bottle of Koelsch on Crafty.
Tom you did misunderstand something here. You have to put the bottle of
Koelsch on *Rebel*. I mean, who will drink all that stuff if Crafty wins
as we all expect it ?!
Robert Hyatt
Peter W. Gillgasch (gil...@ilk.de) wrote:
: I agree. Nevertheless it is a brave thing to do for Ed. I mean, if it
: will be the massacre most of us are expecting it takes a man to post
: those games on his commercial web site.
I don't think it takes bravery here at all, because even if he loses
10-0, it doesn't mean a thing, other than he would also lose by that
or worse against Deep Blue. It doesn't mean a thing about Crafty
being better or worse than Rebel, of course, because 100X is a truly
*huge* computing advantage.
The other match will be more interesting from a purely pragmatic point
of view (He's going to run a *normal* match after (/during?) the
humongous match, just to provide us with information about how Crafty
does on equal timing. If Crafty was to lose 2 for every one it wins
at the equal match, but still win big in the handicap match, it lends
even more credibility to the speed difference issue. The only bad
outcome, which is not likely, is that should Crafty win 2 for every
1 it loses at equal time controls, then the big match would not mean
much. As I said, very unlikely. I'm curious about both parts of this
myself...
Bob
Peter W. Gillgasch
Moritz Berger <Moritz...@msn.com> wrote:
[ Koelsch ]
> Some people say its more a kind of lemonade with alcohol ...
With alcohol ? Are you sure ? Really ? ;)
> It's
> usually served in long, thin glasses in a quantity of 0,2l.
>
> Wow, now I will be flamed ... ;-))))
No you won't. Basically you are right. It *is* lemonade 8^) But
sometimes lemonade is ok too 8^)
Robert Hyatt
Albert Silver (ka...@embratel.net.br) wrote:
: Although I'm not arguing that Rebel will win the handicapped match, I
: don't think the match giving Mchess 2 seconds is really significant to
: disprove Ed. As I recall, Bob told us that when your machine is
: calculating 14-16 plys every move, adding another couple of plys isn't
: going to give the machine a huge difference in strength. Though when you
: go from 4 plys to 6, then it IS significant. That's why Ed only accepted
: the wager if Rebel could think for 3-4 minutes at which point it will see
: quite a bit, and he doesn't think that just adding a few plys will give a
: weaker program (no offence Bob) a crushing edge. Still, though it is
: interesting, I don't think that this will prove anything as far as Deep
: Blue is concerned because as Bob pointed out, it suffers NO slowdown no
: matter how many algorithms it uses, and that is definitely not the case
: for programs running on microcomputers. Getting back to the Mchess match:
: give IT 2-3 minutes and the other the same proportionate advantage in
: time, and see if it still only manages to draw the match (you can even
: repeat the openings), and I'll bet heavily in favor of Mchess.
: Albert Silver
Your idea becomes *very* difficult to test. I tried this with the
famous Cray Blitz vs Genius match a few years ago. The problem was, to
let the fast program reach a normal time control /depth, the slow program
ends up getting forever. It's difficult for me to sit and watch a game
that takes 48 hours to complete (remember, genius is manual so it takes a
human.) In the case of Rebel/Crafty, it's all automatic, so we can really
pick any sort of experiment we'd like, start it up, and check in on the
progress weekly. At least it's not unbearably tedious to make this
happen.
I've tried some tests like you suggest with my old Mach III, but jeez...
if you give Crafty 3 minutes on a P6, you have to give that old 16mhz
68000 forever. talk about getting bored. :)
We will learn two important pieces of information from "the experiment"
however...
1. how Rebel and Crafty compare (something likely only interesting to
me of course... or to those fooling around with Crafty or helping me test
and modify it.)
2. how much 3-4 plies are worth. IE, if the even-up match has them playing
reasonably close to each other strength wise, and the 3-4 ply handicap match
doesn't result in an overwhelming margin of victory for Crafty, then I'll
likely re-think what I'm doing, particularly with the idea of making it
faster on parallel machines and so forth. And try to stick at the current
10-12 plies in the middlegame, and use new speed for more eval, more
extensions, or whatever. If the 3-4 plies does produce a signficant
margin of victory, as compared to the equal-time match, then I'm going to
continue to work on speed/depth and add knowledge as I go, so long as it
doesn't dominate the speed issue too badly.
Be interesting...
Robert Hyatt
Albert Silver (ka...@embratel.net.br) wrote:
: In article <5fvqoc$r...@juniper.cis.uab.edu>,
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
: Albert Silver
Can you post it in PGN? Or at least a reasonable set of moves, and I'll
fake the PGN headers myself. We can give it a whirl. How about any IM's
that are familiar with this game... good candidate?? We really want a
game that is as tactics-free as possible, so it's all strategic. We know
the machines will pick up on the tactics... but the issue is, when searching
*very* deep, does the extra depth produce anything of strategic value, or is
the only gain tactical??
mclane
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>Tom C. Kerrigan (kerr...@merlin.pn.org) wrote:
>: Peter W. Gillgasch (gil...@ilk.de) wrote:
>: > Ed Schroder <rebc...@xs4all.nl> wrote:
>: > > Make your bets now!
>: > Crafty will surely kill Rebel at this time odds match.
>: > I am betting one bottle of Bavarian beer. Anybody ?
>: > -- Peter
>: Eew. I'll put a bottle of Koelsch on Crafty.
>: Cheers,
>: Tom
>I'm going to regret this, but what the hell is a bottle of koelsch?
>:)
Ha ! American meets german beer !!! Here in germany we have many types
of beer. Maybe Peter Gillgasch (as an expert for beer) knows HOW MANY
different beers we have.
We have also a special law that says that in a beer only natural
ingredients have to be used. So NO chemical or whatever stuff.
Koelsch is a beer that is used in the area of cologne.
Can Peter or Tom explains us the different taste of Koelsch in
relation to others ?!
Robert Hyatt
: [ Koelsch ]
: -- Peter
Personally, I don't drink *anything* that is yellow when drunks are
anywhere around. :)
Peter W. Gillgasch
Robert Hyatt <hy...@crafty.cis.uab.edu> wrote:
> Peter W. Gillgasch (gil...@ilk.de) wrote:
>
> : I agree. Nevertheless it is a brave thing to do for Ed. I mean, if it
> : will be the massacre most of us are expecting it takes a man to post
> : those games on his commercial web site.
>
> I don't think it takes bravery here at all, because even if he loses
> 10-0, it doesn't mean a thing, other than he would also lose by that
> or worse against Deep Blue. It doesn't mean a thing about Crafty
> being better or worse than Rebel, of course, because 100X is a truly
> *huge* computing advantage.
But it is still a dangerous thing to do. I am sure that many potential
customers don't understand the technical implications of 100:1 time
odds, let alone the scientific idea behind the match. Will peek at Ed's
website tonite to see how he presents this...
> The other match will be more interesting from a purely pragmatic point
> of view (He's going to run a *normal* match after (/during?) the
> humongous match, just to provide us with information about how Crafty
> does on equal timing. If Crafty was to lose 2 for every one it wins
> at the equal match, but still win big in the handicap match, it lends
> even more credibility to the speed difference issue. The only bad
> outcome, which is not likely, is that should Crafty win 2 for every
> 1 it loses at equal time controls, then the big match would not mean
> much. As I said, very unlikely. I'm curious about both parts of this
> myself...
Didn't realize that you both want do to a "calibration" match after the
100:1 time odds match. Great idea IMHO. Since Rebel didn't compete in CC
events for some time (last time was the wccc in 95 I think) it will be
interesting in any case... I will not bet any beers on the outcome of
that 2nd match of course 8^)
Robert Hyatt
Me either... although I'd probably lean toward Rebel 8 if the stakes were high
enough and I *had* to pick a side. :)
Peter W. Gillgasch
mclane <mcl...@prima.ruhr.de> wrote:
> hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
[ snip ]
> >I'm going to regret this, but what the hell is a bottle of koelsch?
Bob, are you saying that you didn't drink one at the world championship
in Cologne ? I surely had my fill with Bavarian beer that was produced
in mainland China in Hong Kong... They licensed it from us 8^)
> Ha ! American meets german beer !!! Here in germany we have many types
> of beer. Maybe Peter Gillgasch (as an expert for beer) knows HOW MANY
> different beers we have.
In the part of Germany where I was born (Frankonia) every village has at
least one brewery. Many have two. 6 houses and a brewery. Go figure 8^)
I think I recall that I have read some years ago that there are more
than 6,000 different German beers, but I am not totally sure. Could be
more.
> We have also a special law that says that in a beer only natural
> ingredients have to be used. So NO chemical or whatever stuff.
500+ years old Bavarian law. Expanded to the whole republic later...
> Koelsch is a beer that is used in the area of cologne.
"Used" is probably the right word. Some days ago a friend brought this
PowerBook 520c (french system software. Yikes!) over and asked me to
install a weird HP printer. But before I did that we "installed" a
couple of beers. His words 8^)
> Can Peter or Tom explains us the different taste of Koelsch in
> relation to others ?!
Worse 8^)
Robert Hyatt
: > hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
: [ snip ]
: > >I'm going to regret this, but what the hell is a bottle of koelsch?
: Bob, are you saying that you didn't drink one at the world championship
: in Cologne ? I surely had my fill with Bavarian beer that was produced
: in mainland China in Hong Kong... They licensed it from us 8^)
Actually I wasn't at Cologne... I had just started work on my Ph.D. here
and couldn't take off for that long. Harry Nelson went, and we operated
Cray Blitz remotely and had it relay the moves to the tourney. So no, I
wasn't there. "twas a pity however... would have loved to made the trip
Harry and his wife had a great time, and spent another couple of weeks touring
Europe (and dragging that 40kg block of plexiglass we won around with them.:) )
:
: > Ha ! American meets german beer !!! Here in germany we have many types
: > of beer. Maybe Peter Gillgasch (as an expert for beer) knows HOW MANY
: > different beers we have.
: In the part of Germany where I was born (Frankonia) every village has at
: least one brewery. Many have two. 6 houses and a brewery. Go figure 8^)
: I think I recall that I have read some years ago that there are more
: than 6,000 different German beers, but I am not totally sure. Could be
: more.
This sounds like one of those signs in the men's room? IE, "please flush
twice, it's a long way to the local brewery." :)
: > We have also a special law that says that in a beer only natural
Rolf Tueschen
Ed Schroder <rebc...@xs4all.nl> wrote:
>The Crafty-Rebel NPS challenge started!!!
>Make your bets now!
>- Ed Schroder -
------------------------------------------------------------------------------------------------------
To all:
Krafti vs Rebel intelligence foreseen as (between) 4-6 and 3-7.
No doubt about it.
Tueschen R. March 11 1997
-----------------------------------------------------------------------------------------------------
mclane
gil...@ilk.de (Peter W. Gillgasch) wrote:
>mclane <mcl...@prima.ruhr.de> wrote:
>> hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>[ snip ]
>> >I'm going to regret this, but what the hell is a bottle of koelsch?
>Bob, are you saying that you didn't drink one at the world championship
>in Cologne ? I surely had my fill with Bavarian beer that was produced
>in mainland China in Hong Kong... They licensed it from us 8^)
What, Bob was in cologne ? World computer-championship ? Where Ed was
strong and Richard not at his best ? Or do you speak of boring human
championships?
>
>> Ha ! American meets german beer !!! Here in germany we have many types
>> of beer. Maybe Peter Gillgasch (as an expert for beer) knows HOW MANY
>> different beers we have.
>In the part of Germany where I was born (Frankonia) every village has at
>least one brewery. Many have two. 6 houses and a brewery. Go figure 8^)
>I think I recall that I have read some years ago that there are more
>than 6,000 different German beers, but I am not totally sure. Could be
>more.
>> We have also a special law that says that in a beer only natural
>> ingredients have to be used. So NO chemical or whatever stuff.
>500+ years old Bavarian law. Expanded to the whole republic later...
>
>> Koelsch is a beer that is used in the area of cologne.
>"Used" is probably the right word. Some days ago a friend brought this
>PowerBook 520c (french system software. Yikes!) over and asked me to
>install a weird HP printer. But before I did that we "installed" a
>couple of beers. His words 8^)
>
Install a couple of beers ?! AhA.
mclane
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>Peter W. Gillgasch (gil...@ilk.de) wrote:
>: mclane <mcl...@prima.ruhr.de> wrote:
>: > hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>: [ snip ]
>: > >I'm going to regret this, but what the hell is a bottle of koelsch?
>: Bob, are you saying that you didn't drink one at the world championship
>: in Cologne ? I surely had my fill with Bavarian beer that was produced
>: in mainland China in Hong Kong... They licensed it from us 8^)
>Actually I wasn't at Cologne... I had just started work on my Ph.D. here
>and couldn't take off for that long. Harry Nelson went, and we operated
>Cray Blitz remotely and had it relay the moves to the tourney. So no, I
>wasn't there. "twas a pity however... would have loved to made the trip
>Harry and his wife had a great time, and spent another couple of weeks touring
>Europe (and dragging that 40kg block of plexiglass we won around with them.:) )
I was at cologne, its a massive time back ? When was it ? 1986 ?
>: > Ha ! American meets german beer !!! Here in germany we have many types
>: > of beer. Maybe Peter Gillgasch (as an expert for beer) knows HOW MANY
>: > different beers we have.
>: In the part of Germany where I was born (Frankonia) every village has at
>: least one brewery. Many have two. 6 houses and a brewery. Go figure 8^)
>: I think I recall that I have read some years ago that there are more
>: than 6,000 different German beers, but I am not totally sure. Could be
>: more.
>This sounds like one of those signs in the men's room? IE, "please flush
>twice, it's a long way to the local brewery." :)
Although this is off-topic, anybody joins it. Aha. We are humans. We
can do more than talk about computerchess. We can talk a lot about
drugs and toilets.
Tom C. Kerrigan
> >: > > Make your bets now!
> >: > Crafty will surely kill Rebel at this time odds match.
> >: > I am betting one bottle of Bavarian beer. Anybody ?
> >: > -- Peter
> >: Eew. I'll put a bottle of Koelsch on Crafty.
> >: Cheers,
> >: Tom
> Ha ! American meets german beer !!! Here in germany we have many types
> of beer. Maybe Peter Gillgasch (as an expert for beer) knows HOW MANY
> different beers we have.
At least 5k microbreweries, I've read...
> We have also a special law that says that in a beer only natural
> ingredients have to be used. So NO chemical or whatever stuff.
Well, it's not a law anymore, thanks to the EC. Most German beers are
still compliant, though.
> Koelsch is a beer that is used in the area of cologne.
Eh? "Used"? You make it sound like we wash our cars with it. :)
> Can Peter or Tom explains us the different taste of Koelsch in
> relation to others ?!
Well, Koelsch looks/tastes more like Pils than Alt. My personal theory is
that Alt is watered down topsoil. :) Anyway, I think it's a bit smoother
than the average Pils, and a bit thicker than my personal favorite, wheat
beer. (Which is my personal favorite because it doesn't taste so bad,
plus it comes in great big cool-shaped glasses.)
BTW, I always get Koelsch from bottles that are a lot bigger than ,2l...
Cheers,
Tom
Robert Hyatt
mclane (mcl...@prima.ruhr.de) wrote:
: gil...@ilk.de (Peter W. Gillgasch) wrote:
: >mclane <mcl...@prima.ruhr.de> wrote:
: >> hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
: >[ snip ]
: >> >I'm going to regret this, but what the hell is a bottle of koelsch?
: >Bob, are you saying that you didn't drink one at the world championship
: >in Cologne ? I surely had my fill with Bavarian beer that was produced
: >in mainland China in Hong Kong... They licensed it from us 8^)
: What, Bob was in cologne ? World computer-championship ? Where Ed was
: strong and Richard not at his best ? Or do you speak of boring human
: championships?
He's speaking of the 1986 World Computer Chess Championship, which was
won by Cray Blitz. No, I wasn't there. Harry Nelson, from Lawrence
Livermore National Laboratory (The nuclear physics guys) was a member
of the Cray Blitz "team of 3" and went there as part of his vacation
that year...
Jack Nerad
Ed Schroder wrote:
>
> The Crafty-Rebel NPS challenge started!!!
>
> Moritz!) and the 1:100 match has started.
>
> The comparison 50-60 game match on equal tournament time (4:30 average)
> will start in a few days too.
>
> You will find the match at:
>
> http://www.xs4all.nl/~rebchess/match.htm
>
> Make your bets now!
>
> - Ed Schroder -
The current eval for Crafty (3/11 at 14:30 EST) is -.188 by Crafty and
.36 by Rebel8. Does this mean that Crafty is losing? Assuming that
Crafty will pull ahead, how long, IYO, will it take before the advantage
becomes apparent?
Jack Nerad
Peter W. Gillgasch
Tom C. Kerrigan <kerr...@merlin.pn.org> wrote:
> mclane (mcl...@prima.ruhr.de) wrote:
> > We have also a special law that says that in a beer only natural
> > ingredients have to be used. So NO chemical or whatever stuff.
>
> Well, it's not a law anymore, thanks to the EC. Most German beers are
> still compliant, though.
Right. There was a *huge* discussion about the admission of foreign
beers in Germany that are not compliant.
> > Koelsch is a beer that is used in the area of cologne.
>
> Eh? "Used"? You make it sound like we wash our cars with it. :)
You don't ?
> Well, Koelsch looks/tastes more like Pils than Alt. My personal theory is
> that Alt is watered down topsoil. :)
Hey, Alt is ok ! You know this Toten Hosen song, don't you ?
> Anyway, I think it's a bit smoother
> than the average Pils, and a bit thicker than my personal favorite, wheat
> beer. (Which is my personal favorite because it doesn't taste so bad,
> plus it comes in great big cool-shaped glasses.)
Cool. We agree on wheat beer being the best. Now the tie breaker:
Do you prefer:
Light wheat beer (looks yellow/golden)
Dark wheat beer (dark brown stuff)
Robert Hyatt
Jack Nerad (JNE...@concentric.net) wrote:
: Jack Nerad
Much too early. -.188 is "chickenfeed" for crafty, which sees positional
scores of +1.5 to -1.5 all the time, and occasionally sees them bounce to
+/- 5.000 at times.
When either gets to +1.000 or -1.000, then the other side will start getting
worried or happy. :) For now, it's just a game...
brucemo
Jack Nerad wrote:
> The current eval for Crafty (3/11 at 14:30 EST) is -.188 by Crafty and
> .36 by Rebel8. Does this mean that Crafty is losing? Assuming that
> Crafty will pull ahead, how long, IYO, will it take before the advantage
> becomes apparent?
I looked at these scores and thought to myself, "Given that it is so early
in the game, I bet that what is going on here is that Rebel is playing
white in a Sicilian Defense", and in fact, that is what is happening.
You can't be ahead all the time in every game, especially if you play black
in some of them.
This is why most of us would prefer to play white.
bruce
ShaktiFire
Well for my two cents worth... and with my chess knowledge
thats about all you get :-).
I think 0.36 advantage is still in the mud, not yet significant.
The game to me looks like a tactical melee in the making
and I would rather slug it out at 14 plies than 9 plies.
Does anyone have this in one of the 1 million game
databases? Probably not, it looks unusual, but could
you all check anyhow?
Best Wishes
Robert Hyatt
brucemo (bru...@nwlink.com) wrote:
: Jack Nerad wrote:
: > The current eval for Crafty (3/11 at 14:30 EST) is -.188 by Crafty and
: > .36 by Rebel8. Does this mean that Crafty is losing? Assuming that
: > Crafty will pull ahead, how long, IYO, will it take before the advantage
: > becomes apparent?
: I looked at these scores and thought to myself, "Given that it is so early
: in the game, I bet that what is going on here is that Rebel is playing
: white in a Sicilian Defense", and in fact, that is what is happening.
: You can't be ahead all the time in every game, especially if you play black
: in some of them.
: This is why most of us would prefer to play white.
: bruce
I just hope to be ahead in *some* of them. :) One *very* difficult lesson
I've learned over 25+ years of doing this is to be careful of the expectations
I have when popping out of book. If you try to maximize your score, chances
are you are coming out in a drawish position, which is what we saw too much
of in the last few games Cray Blitz played... I've since modified my "wants"
to now reach positions where the score is negative, but there is some inklings
of compensation for that - score. IE, unbalanced positions, where Crafty (or
any program) has a chance to play... If you watch crafty, you'll notice that
it is almost always - coming out of book, but the position is unbalanced enough
that (against people) it has something to struggle with. I don't want positions
where (say) my opponent has an isolated queen-pawn, but there's *nothing*
else going on, because that is an easier position to play for both sides. If
there are unbalanced aspects of the position, you have to commit to one or the
other, usually. That gives chances for something good (and something bad at
times of course) to happen...
mark lefler
> Ed Schroder wrote:
> >
> > The Crafty-Rebel NPS challenge started!!!
Great web spot showing Rebel's analysis. Ed, could you also post Crafts PV
and analysis?
Oddly, crafty has so far only matched one of Rebels predicted moves.
For what it's worth, I put the same positions in my program NOW and it
agrees rather closely with Rebel on all but one move (which was a
transposition). NOW's score are also much closer to Rebels so far.
This will be a very interesting match!
-->Mark Lefler
Robert Hyatt
mark lefler (72626...@compuserve.com) wrote:
: -->Mark Lefler
Don't forget the depth difference. That can really change the prediction
and scores a significant amount. the only funny part here is that the poor
hash table has 384,000 entries. at 100K nps, this works out to a saturation
factor that has only been the subject of nightmares, since crafty's searching
about 36,000 seconds, that works out to be some 3,600,000,000 nodes per
move, or about 10000X saturation. Good thing ed's machine has DRAM... it
would melt magnetic cores.. :)
It's certainly stress-testing my replacement policy. :)
Ed Schroder
From: "mark lefler" <72626...@compuserve.com>
: Great web spot showing Rebel's analysis. Ed, could you also post
: Crafts PV and analysis?
I created the analyze text with Rebel's option "Game to Textfile".
I discovered a feature in Rebel8 I wasn't even aware of... :))
- Game to Textfile;
- a:game1.txt
Rebel8 analyse on flop ready to port to the Internet without leaving
the program.
I noticed this trick also works for Crafty (using a:) for file output.
But perhaps Bob knows the right command to get Crafty's analyze this way?
After game-1 I will make Crafty's complete logfile available anyway.
Rolf Czedzak
Peter W. Gillgasch wrote: <199703090448542055741@[194.121.104.134]>
PWG> Ed Schroder <rebc...@xs4all.nl> wrote:
PWG>
PWG> > Make your bets now!
PWG>
PWG> Crafty will surely kill Rebel at this time odds match.
PWG>
PWG> I am betting one bottle of Bavarian beer. Anybody ?
Bavarian beer -looks like You're not very confident about the outcome
of the match. Or do You just want to get rid of some stuff? ;-))
PWG> -- Peter
Rolf C
Rolf Czedzak
Robert Hyatt wrote: <5g1ku2$j...@juniper.cis.uab.edu>
RH> Peter W. Gillgasch (gil...@ilk.de) wrote:
RH>
RH> : May God grant me the serenity to accept the things I cannot change,
RH> : courage to choke the living shit out of those who piss me off,
RH> : and wisdom to know where I should hide the bodies...
RH>
RH> Personally, I don't drink *anything* that is yellow when drunks are
RH> anywhere around. :)
Don't worry, Peter's sig isn't directly related to "Koelsch", although ...
;-)
Rolf C
Rolf Czedzak
Peter W. Gillgasch wrote: <19970310154828441536@[194.121.104.133]>
PWG> Tom C. Kerrigan <kerr...@merlin.pn.org> wrote:
PWG>
PWG> > Eew. I'll put a bottle of Koelsch on Crafty.
PWG>
PWG> At least one guy with a sense for the important things
PWG> in life on this newsgroup...
PWG>
PWG> Bring the bottle to the next world micro. Bring some for yourself,
PWG> you will need it 8^)
You already drunk Your bottle(s) of beer?? Or how comes You can't get
right who's betting on what outcome? ;-)))
Rolf Czedzak
Robert Hyatt wrote: <5fvq3m$r...@juniper.cis.uab.edu>
RH> I'm going to regret this, but what the hell is a bottle of koelsch?
RH>
RH> :)
Some fluid, they call beer in the Cologne region. Blonds use it to
wash their hair.
Rolf C
Albert Silver
In article <19970312005...@ladder01.news.aol.com>,
shakt...@aol.com (ShaktiFire) wrote:
>
> Well for my two cents worth... and with my chess knowledge
> thats about all you get :-).
>
> I think 0.36 advantage is still in the mud, not yet significant.
> The game to me looks like a tactical melee in the making
> and I would rather slug it out at 14 plies than 9 plies.
>
Actually, what might be interesting would be to see what Rebel thinks
when IT reaches that depth. Sure, for the match it's only going a few
minutes, but for analytical purposes, would it change it's moves or
evaluation significantly if it were to go that deep in it's calculations?
Ed?
Harald Faber
quoting a mail from mclane # prima.ruhr.de
Hello mclane,
m> From: mcl...@prima.ruhr.de (mclane)
m> Subject: Re: The Crafty-Rebel NPS challenge started!!!
m> Organization: Prima e.V. Dortmund
m> >: > Crafty will surely kill Rebel at this time odds match.
m> >: > I am betting one bottle of Bavarian beer. Anybody ?
m> >: > -- Peter
m> >: Eew. I'll put a bottle of Koelsch on Crafty.
m> >: Tom
m> >I'm going to regret this, but what the hell is a bottle of koelsch?
m> >:)
m> Ha ! American meets german beer !!! Here in germany we have many types
m> of beer. Maybe Peter Gillgasch (as an expert for beer) knows HOW MANY
m> different beers we have.
m> We have also a special law that says that in a beer only natural
m> ingredients have to be used. So NO chemical or whatever stuff.
m> Koelsch is a beer that is used in the area of cologne.
m> Can Peter or Tom explains us the different taste of Koelsch in
m> relation to others ?!
Better ask Podz-Blitz (K.-H. Podzielny) at next tournament at short time
controls, he knows almost everything about beer and will laugh at you for
this "...no chemical stuff..."
He always tells you that he ordered test reports about German beers from
the Stiftung Warentest every month. He can tell you which beers are otu of
chemical stuff.... :-)
Harald
--
Harald Faber
quoting a mail from gillga # ilk.de
Hello Peter,
PWG> From: gil...@ilk.de (Peter W. Gillgasch)
PWG> Subject: Re: The Crafty-Rebel NPS challenge started!!!
PWG> Organization: Organised ? Who ? Me ?
PWG> Ed Schroder <rebc...@xs4all.nl> wrote:
PWG> > Make your bets now!
PWG> Crafty will surely kill Rebel at this time odds match.
PWG> I am betting one bottle of Bavarian beer. Anybody ?
PWG> -- Peter
I would take Bitburger, Flens or Jever, no compromise. :-)
Harald
--
Amir Ban
Tom C. Kerrigan wrote:
[intensive alcohol discussion snipped]
> Cheers,
> Tom
Why is it, whenever a r.g.c.c thread becomes really busy, you can bet
that beer or beer-cans are being discussed ?
Amir
Tom C. Kerrigan
Peter W. Gillgasch (gil...@ilk.de) wrote:
> > Anyway, I think it's a bit smoother
> > than the average Pils, and a bit thicker than my personal favorite, wheat
> > beer. (Which is my personal favorite because it doesn't taste so bad,
> > plus it comes in great big cool-shaped glasses.)
> Cool. We agree on wheat beer being the best. Now the tie breaker:
> Do you prefer:
> Light wheat beer (looks yellow/golden)
> Dark wheat beer (dark brown stuff)
Light, but only a bit over Dark...
For worst beer, I'd have to go with Diebels... Uff...
Cheers,
Tom
mclane
Amir Ban <ami...@msys.co.il> wrote:
>Amir
Because we are humans ! Computers don't drink! And don't buy cans of
beer.
Harald Faber
quoting a mail from hyatt # crafty.cis.uab.edu
Hello Robert,
RH> From: hy...@crafty.cis.uab.edu (Robert Hyatt)
RH> Subject: Re: The Crafty-Rebel NPS challenge started!!!
RH> Organization: CIS, University of Alabama at Birmingham
RH> : > Good idea. Here is *your* mission. Find a neat positional game. Not
RH> : > one with a lot of tactical fireworks. IE, find a karpov vs someone
RH> : > game. : bruce
RH> : How about Fischer - Reschevsky US Championship 1962-3 (game no.43 in his
RH> : 60 memorable games)? That game always struck me as being an absolute
RH> : masterpiece in position play and clarity.
RH> : Albert Silver
RH> Can you post it in PGN? Or at least a reasonable set of moves, and I'll
RH> fake the PGN headers myself. We can give it a whirl. How about any IM's
RH> that are familiar with this game... good candidate?? We really want a
RH> game that is as tactics-free as possible, so it's all strategic. We know
RH> the machines will pick up on the tactics... but the issue is, when
RH> searching *very* deep, does the extra depth produce anything of strategic
RH> value, or is the only gain tactical??
Ok it is here, look at subject "Fischer-Reschevsky"
Harald
--
Martin Zentner
mclane wrote:
>
> Amir Ban <ami...@msys.co.il> wrote:
[... history of posters deleted ...]
> >[intensive alcohol discussion snipped]
> >
> >> Cheers,
> >> Tom
>
> >Why is it, whenever a r.g.c.c thread becomes really busy, you can bet
> >that beer or beer-cans are being discussed ?
>
> >Amir
>
> Because we are humans ! Computers don't drink! And don't buy cans of
> beer.
Anyway, it would be nice, if people would change the subject, when
they start discussing off-topic. This would make it easier to
read a large newsgroup like r.g.c.c.
As long as it is somehow related to computer chess, I wouldn't want
to miss those discussions, but please use a new topic !
I would certainly like to know, who's getting pissed, when, why, etc.
But normally I would not expect this to be the 100th repply to
an old announcement. Therefore this is likely to be missed. :-(
And you can't expect, that everyone has the time to read every
article posted here. (ok, Bob maybe has the time :-)
Do not just hit the reply button and start typing without changing the
subject, although this seems logical. Maybe it's even possible to edit
some of the original posting, when including it. Please, keep it short
and to-the-point. Especially Bob includes everything from previous
postings although just referencing few points.
Thanks
-Martin
--
Martin Zentner, University of Passau [http://www.uni-passau.de/]
E-Mail: mailto:zen...@phil.uni-passau.de
WWW: http://www.phil.uni-passau.de/linguistik/staff/zentner/
Tom C. Kerrigan
Amir Ban (ami...@msys.co.il) wrote:
> Why is it, whenever a r.g.c.c thread becomes really busy, you can bet
> that beer or beer-cans are being discussed ?
Because [some] beer is pretty good stuff.
It's not like such discussions suck enough bandwidth to preclude
conversations about, say, hash table replacement schemes.
Cheers,
Tom
mclane
Martin Zentner <zen...@phil.uni-passau.de> wrote:
>> Because we are humans ! Computers don't drink! And don't buy cans of
>> beer.
>Anyway, it would be nice, if people would change the subject, when
>they start discussing off-topic. This would make it easier to
>read a large newsgroup like r.g.c.c.
>As long as it is somehow related to computer chess, I wouldn't want
>to miss those discussions, but please use a new topic !
Sorry Martin.
Tim Mirabile
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>Albert Silver (ka...@embratel.net.br) wrote:
>: In article <5fvqoc$r...@juniper.cis.uab.edu>,
>: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>
>
>: > Good idea. Here is *your* mission. Find a neat positional game. Not one
>: > with a lot of tactical fireworks. IE, find a karpov vs someone game.
>: > : bruce
>
>: How about Fischer - Reschevsky US Championship 1962-3 (game no.43 in his
>: 60 memorable games)? That game always struck me as being an absolute
>: masterpiece in position play and clarity.
>
>: Albert Silver
>
>: -------------------==== Posted via Deja News ====-----------------------
>: http://www.dejanews.com/ Search, Read, Post to Usenet
>
>Can you post it in PGN? Or at least a reasonable set of moves, and I'll
>fake the PGN headers myself. We can give it a whirl. How about any IM's
>that are familiar with this game... good candidate?? We really want a
>game that is as tactics-free as possible, so it's all strategic. We know
>the machines will pick up on the tactics... but the issue is, when searching
>*very* deep, does the extra depth produce anything of strategic value, or is
>the only gain tactical??
[Event "?"]
[Site "ch-USA"]
[Date "1962.??.??"]
[Round "?"]
[White "Fischer,R"]
[Black "Reshevsky,S"]
[Result "1-0"]
[ECO "B90"]
1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. h3 g6 7. g4 Bg7 8. g5 Nh5
9. Be2 e5 10. Nb3 Nf4 11. Nd5 Nxd5 12. Qxd5 Nc6 13. Bg4 Bxg4 14. hxg4 Qc8 15.
Qd1 Nd4 16. c3 Nxb3 17. axb3 Qe6 18. Ra5 f6 19. Qd5 Qxd5 20. Rxd5 Kd7 21. gxf6
Bxf6 22. g5 Be7 23. Ke2 Raf8 24. Be3 Rc8 25. b4 b5 26. Rdd1 Ke6 27. Ra1 Rc6 28.
Rh3 Bf8 29. Rah1 Rc7 30. Rh4 d5 31. Ra1 Rc6 32. exd5+ Kxd5 33. Rd1+ Ke6 34. Rd8
Kf5 35. Ra8 Re6 36. Rh3 Bg7 37. Rxh8 Bxh8 38. Rxh7 Re8 39. Rf7+ Kg4 40. f3+ Kg3
41. Kd3 e4+ 42. fxe4 Rd8+ 43. Bd4 Kg4 44. Rf1 Be5 45. Ke3 Bc7 46. Rg1+ Kh4 47.
Kf3 Rd7 48. e5 Rf7+ 49. Ke4 Rf5 50. e6 Bd8 51. Bf6 Bxf6 52. gxf6 Rxf6 53. Kd5
Rf2 54. Re1 1-0
We have Fischer's very detailed notes to go by, so perhaps this is a good game
to look at.
--
Tim Mirabile <t...@mail.htp.com> - http://www.webcom.com/timm/
Visit my homepage for information on USCF & FIDE rated chess on Long Island.
TimM on the Free Internet Chess Server - telnet://fics.onenet.net:5000/
Webmaster, tech support - ICD/Your Move Chess & Games: http://www.icdchess.com/
The opinions of my employers are not necessarily mine, and vice versa.
Jack Nerad
The point of the match is to see how many points stronger a program
becomes given greater amount of search time, correct? The series of
games played after the ten game time-odds match will be used as a
reference for judgement of this point. My question is this: will only
ten games in the time-odds match provide an accurate enough measure of
this statistically?
Jack Nerad
Robert Hyatt
Jack Nerad (JNE...@concentric.net) wrote:
: The point of the match is to see how many points stronger a program
: Jack Nerad
possibly not, but the obvious problem is the games are going to take a long
time to play. As in weeks per game. If the experiment doesn't end pretty
quickly, it will span a time-frame that covers multiple rebel and Crafty
revisions. If the split is 8-2, the extra depth will be pretty convincing.
If it is 2-8, then it will say that Rebel's evel is much better than Crafty's.
If it is anything like 5-5 it will be inconclusive. Game one is interesting.
It was played with Crafty not having a book, and Crafty went into this g6
variation of the Sicilian on its own. It then followed that with the c4
stuff to open the position up a little. And finally, it ended up with a
passed e-pawn that may or may not mean anything in this game. Unfortunately,
passed pawns are *very* valuable in Crafty's evaluation, and it found a way to
make one, at the expense of exposing itself to an attack of sorts. So, the
game is interesting. I'd bet that if the timing were equal, that Rebel would
probably win this game handily because of the attacking potential for white.
*but*...Crafty is searching very deeply, and white's king is not all that safe
either. After the expected Rb6 and Nd8 both sides have to walk a thin line
for a while. White probably will try Bc1 to prevent Rh6 which pins the white
queen on the king.
In short, it's an interesting game. Certainly not the typical boring
"computer" games we see all the time...
As to whether we can conclude anything about the speed difference, we'll just
have to see how the first few games go. One person has already gone on the
podium and claimed that Rebel's going to win the match even with the handicap.
If Rebel doesn't, I hope that poster will fade away, since he relies more on
emotion than on fact.
You can run this test yourself if you have Crafty and xboard/winboard...
try
xboard -mm -fcp "crafty xboard st=30" -scp "crafty xboard st=300" and see
what happens. 10:1 time odds. I do this a lot, and the 10-1 odds is an
overwhelming advantage...
Jack Nerad
Robert Hyatt wrote:
>
> Jack Nerad (JNE...@concentric.net) wrote:
> : The point of the match is to see how many points stronger a program
> : becomes given greater amount of search time, correct? The series of
> : games played after the ten game time-odds match will be used as a
> : reference for judgement of this point. My question is this: will only
> : ten games in the time-odds match provide an accurate enough measure of
> : this statistically?
>
> : Jack Nerad
>
> possibly not, but the obvious problem is the games are going to take a long
> time to play. As in weeks per game. If the experiment doesn't end pretty
> quickly, it will span a time-frame that covers multiple rebel and Crafty
> revisions. If the split is 8-2, the extra depth will be pretty convincing.
> If it is 2-8, then it will say that Rebel's evel is much better than Crafty's.
> If it is anything like 5-5 it will be inconclusive. Game one is interesting.
mistake. If Crafty is 400 points weaker than Rebel and gains 400 points
thanks to the time-odds, you wouldn't know. The idea to run the
reference matches is an excellent one for exactly this reason: you get a
better idea of Crafty's strength at normal time odds.
My suggestion is that even if Crafty wins one game in the time-odds
match, but loses every game in the reference series, that there must be
some advantage in the greater time allotment. Since the win would
indicate an increase in the playing strength. The amount of the
increase would be difficult to determine, however, if Crafty loses every
game in the reference match because the "Elo" difference could be 400 or
it could be 800, without wins, it is impossible to say.
In a ten game match, as well, the win percentages are limited to
multiples of ten, which, while it will prove that strength=speed, may or
may not be satisfactory in measuring the increase in the performance the
added time allows.
> As to whether we can conclude anything about the speed difference, we'll just
> have to see how the first few games go. One person has already gone on the
> podium and claimed that Rebel's going to win the match even with the handicap.
It may be that Rebel will win the match, the important question to
answer is: did Crafty's strength increase due to the increased amount of
time it spent solving the problem? and if so, how much did it increase?
as it is these questions which will answer the original controversy of
whether or not searching deeper and deeper will generate a stronger
program.
Jack Nerad
ShaktiFire
Robert Hyatt wrote:
"If it is anything like 5 - 5, it will be inconclusive."
Well I am sorry.... I don't think this is a fair statement.
Originally , though I have not the post, Prof. Hyatt certainly said
that the DB team indicated to him to they were beating PCs 90%.
If my understanding is correct, Bob eventually stated or implied that
if Crafty were given the same ply depth advantage that it would
do a similar result.
Further... Ed stated that this did not seem
correct to him. He admitted the superiority of DB because of its
speed advantage... but he indicated that he felt such a depth advantage
would only give a 65% advantage. This seems to me, the position he
evolved to.(I remember this clearly, and can find DejaNews if need be).
Therefore, a 5 - 5 result is very substantial for
Ed's claim. If the 60 game standard match ends even,
then a 5 - 5 result would logically indicate no increased advantage from
100 - 1 speed. Of course Ed never claimed there would be NO advantage
from speed. He simply claimed there is a diminishing return, and that
the advantage IS NOT 90% but more like 65% in a 10 game match.
If in the "sixty game even time" match the results are disparate in
favor of
Rebel, then it is harder to gauge the results of the ten game match, since
Rebel would then seem to have a better positional eval, and we have a
case of "better eval" vs. "better speed". This is fuzzier to think about
and more testing would need to be done to understand this.
In summary, if Rebel can get 4 - 5 points, it would be decisive idea
victory
for Ed. If Crafty can get 8 - 9 points , it would idea victory for Bob.
6 - 7 points is inconclusive.
That is the simplified idea. Of course it is only vaguely scientific,
since
the uncertainty margins of a 10 game match are huge. But it is a quite
interesting study since it helps us understand whether computer speed
increases will eventually help or not.
I admire Ed for running this test and posting results candidly. I
admire Bob for the wonderful work he has done and is doing
and for sharing his ideas and knowledge with us. So I hope
in some sense, this is the posting of an impartial fan of both. :-)
Best Wishes
Kevin James Begley
Robert Hyatt (hy...@crafty.cis.uab.edu) wrote:
: <snip>
: One person has already gone on the
: podium and claimed that Rebel's going to win the match even with the handicap.
: emotion than on fact.
A common flaw in humans, though sometimes emotion is a better guide.
Kevin.
Robert Hyatt
Jack Nerad (JNE...@concentric.net) wrote:
: Robert Hyatt wrote:
: >
: > Jack Nerad (JNE...@concentric.net) wrote:
: > : The point of the match is to see how many points stronger a program
: > : becomes given greater amount of search time, correct? The series of
: > : games played after the ten game time-odds match will be used as a
: > : reference for judgement of this point. My question is this: will only
: > : ten games in the time-odds match provide an accurate enough measure of
: > : this statistically?
: >
: > : Jack Nerad
: >
: > possibly not, but the obvious problem is the games are going to take a long
: > time to play. As in weeks per game. If the experiment doesn't end pretty
: > quickly, it will span a time-frame that covers multiple rebel and Crafty
: > revisions. If the split is 8-2, the extra depth will be pretty convincing.
: > If it is 2-8, then it will say that Rebel's evel is much better than Crafty's.
: Judging solely by the outcome of the time-odds match seems to be a
: mistake. If Crafty is 400 points weaker than Rebel and gains 400 points
: thanks to the time-odds, you wouldn't know. The idea to run the
: reference matches is an excellent one for exactly this reason: you get a
: better idea of Crafty's strength at normal time odds.
The reference match should be interesting as well, of course. Personally
I don't think that Crafty is 400 points worse than *any* program, except
possibly for Deep Blue's huge speed advantage. I now have literally
dozens of games against most of the current crop of programs, including
40+ against Hiarcs 6. The surprising result? CM5000 is *still* the
strongest opponent (from crafty's perspective). I haven't spent a lot of
time studying these games, so I can's explain why this is true. But the
numbers support this. (BTW, this is only for the commercial programs...
I still think Ferret is the best there is...)
: My suggestion is that even if Crafty wins one game in the time-odds
: match, but loses every game in the reference series, that there must be
: some advantage in the greater time allotment. Since the win would
: indicate an increase in the playing strength. The amount of the
: increase would be difficult to determine, however, if Crafty loses every
: game in the reference match because the "Elo" difference could be 400 or
: it could be 800, without wins, it is impossible to say.
I hope it's not that lopsided. Otherwise I might go back to operating
system programming. :)
: In a ten game match, as well, the win percentages are limited to
: multiples of ten, which, while it will prove that strength=speed, may or
: may not be satisfactory in measuring the increase in the performance the
: added time allows.
My assumption is that the normal match will be "close enough" (even if
Crafty only wins 10 out of 40, which would be a bad result in my eyes)
the time handicap match will be most revealing. I'm a little surprised
by the search depth difference the factor of 100 is giving. I was assuming
about 3 plies... but it is really turning out to be 5 plies generally,
according to the partial results we're seeing from Ed. He has a huge
problem in getting information from either of the two programs while they
are running. I've just added a command that lets him copy part of crafty's
log file to the a: drive so he can cut and paste the analysis for the next
game so we can see scores, PV's and so forth which will give a lot of
insight into what that 5-ply advantage means...
: > As to whether we can conclude anything about the speed difference, we'll just
: > have to see how the first few games go. One person has already gone on the
: > podium and claimed that Rebel's going to win the match even with the handicap.
: It may be that Rebel will win the match, the important question to
: answer is: did Crafty's strength increase due to the increased amount of
: time it spent solving the problem? and if so, how much did it increase?
: as it is these questions which will answer the original controversy of
: whether or not searching deeper and deeper will generate a stronger
: program.
: Jack Nerad
It certainly might happen. Of course, it would obviously *not* speak very
well of Crafty. Nor anyone else for that matter, since Crafty played several
very good programs in Jakarta and won most of the games... But you have to
go figure anyway. For example, Crafty has blown MacChess out in the
Korrespondence Kup, yet Rebel/MacChess is an apparent draw. That alone
leads me to believe that while Rebel might well be better than Crafty, it
is not a huge difference...
Robert Hyatt
ShaktiFire (shakt...@aol.com) wrote:
: Robert Hyatt wrote:
: "If it is anything like 5 - 5, it will be inconclusive."
: Well I am sorry.... I don't think this is a fair statement.
: Originally , though I have not the post, Prof. Hyatt certainly said
: that the DB team indicated to him to they were beating PCs 90%.
: If my understanding is correct, Bob eventually stated or implied that
: if Crafty were given the same ply depth advantage that it would
: do a similar result.
: Further... Ed stated that this did not seem
: correct to him. He admitted the superiority of DB because of its
: speed advantage... but he indicated that he felt such a depth advantage
: would only give a 65% advantage. This seems to me, the position he
: evolved to.(I remember this clearly, and can find DejaNews if need be).
: Therefore, a 5 - 5 result is very substantial for
: Ed's claim. If the 60 game standard match ends even,
: then a 5 - 5 result would logically indicate no increased advantage from
: 100 - 1 speed. Of course Ed never claimed there would be NO advantage
: from speed. He simply claimed there is a diminishing return, and that
: the advantage IS NOT 90% but more like 65% in a 10 game match.
Not necessarily, because there are *two* variables in this equation.
If the test were Crafty at depth=10 vs Crafty at depth=15, there would
be only *one* variable, the depth. Or if we tested Rebel at depth=10
vs Rebel at depth=15, there would be only one variable. Here there are
two. It doesn't take much to make the outcome of Prog.a vs Prog.b change
in a significant way.
However, 5-5 would simply mean that Crafty's eval is too much worse than
Ed's to make the result useful. I don't think that will happen, nor any-
thing close to that, but in this day and time of course, *anything* can
happen. Don't forget that Fritz thumped Deep Blue Prototype in Hong Kong,
something that is *not* likely to happen again.
: If in the "sixty game even time" match the results are disparate in
: favor of
: Rebel, then it is harder to gauge the results of the ten game match, since
: Rebel would then seem to have a better positional eval, and we have a
: case of "better eval" vs. "better speed". This is fuzzier to think about
: and more testing would need to be done to understand this.
: In summary, if Rebel can get 4 - 5 points, it would be decisive idea
: victory
: for Ed. If Crafty can get 8 - 9 points , it would idea victory for Bob.
: 6 - 7 points is inconclusive.
: That is the simplified idea. Of course it is only vaguely scientific,
: since
: the uncertainty margins of a 10 game match are huge. But it is a quite
: interesting study since it helps us understand whether computer speed
: increases will eventually help or not.
: I admire Ed for running this test and posting results candidly. I
: admire Bob for the wonderful work he has done and is doing
: and for sharing his ideas and knowledge with us. So I hope
: in some sense, this is the posting of an impartial fan of both. :-)
I don't mind compliments, I don't mind someone pointing out something
horrible I've done in Crafty. So I don't take offense at most things
posted here. The only difficulty I find myself in is that when I post
something about "Crafty found this in X seconds" I can't fudge. Because
any of about 100,000 different people can respond with "wait just a minute,
I have 11.18 and it takes 2 hours to find that..." You can therefore trust
what I say about it, whether it is good or bad.
BTW for those interested... In cleaning up files on our ftp machine, I took
the ftp log, and with a little work, sorted it, eliminated duplicates, and
found that there have been downloads to over 100,000 *different* hosts.
Which is remarkable. My sysadmin noted that there is almost *always* 2-3
ftp sessions going on downloading crafty. Many are duplicates as people
get the most recent version, but that >100,000 number was surprising.
I have no idea how "understated" that is, because many different people
could download from the same compuserve (say) host, and I'd count that as
"1" for this count... It does make my daily dose of 30 emails seem pretty
reasonable. :)
: Best Wishes
brucemo
Jack Nerad wrote:
>
> Robert Hyatt wrote:
> >
> > Jack Nerad (JNE...@concentric.net) wrote:
> > : The point of the match is to see how many points stronger a program
> > : becomes given greater amount of search time, correct? The series of
> > : games played after the ten game time-odds match will be used as a
> > : reference for judgement of this point. My question is this: will only
> > : ten games in the time-odds match provide an accurate enough measure of
> > : this statistically?
No.
If you run a 10-game match, assuming 40% drawing probability, and the opponents are
exactly the same strength, the match score will be either 40%, 45%, 50%, 55%, or
60%, only about 70% of the time. This means that 30% of the time one of the
programs will score 65% or better, which is at least 100 rating points indicated
difference.
So if we get a score of 65% for one side or the other, I don't think we can even
say that one is stronger than the other. With scores larger than this, maybe you
can say that one is stronger than the other with some degree of confidence, but if
you want to say "X is stronger than Y by Z Elo points", your degree of confidence
drops as you increase Z.
Some folks will love or hate a program based upon the results of a two-game match,
but this is insane. A ten-game match seems like it would be better, but it isn't
much better really, unless the match is a complete wipeout. Then you might be able
to say that one of the programs is probably a little better than the other one.
> > possibly not, but the obvious problem is the games are going to take a long
> > time to play. As in weeks per game. If the experiment doesn't end pretty
> > quickly, it will span a time-frame that covers multiple rebel and Crafty
> > revisions. If the split is 8-2, the extra depth will be pretty convincing.
> > If it is 2-8, then it will say that Rebel's evel is much better than Crafty's.
> > If it is anything like 5-5 it will be inconclusive. Game one is interesting.
If you get a score of 80% in a ten-game match, the program that scored 80% is
probably at least a few Elo points better than the other one given the constraints
of the match.
> Judging solely by the outcome of the time-odds match seems to be a
> mistake. If Crafty is 400 points weaker than Rebel and gains 400 points
> thanks to the time-odds, you wouldn't know. The idea to run the
> reference matches is an excellent one for exactly this reason: you get a
> better idea of Crafty's strength at normal time odds.
I don't know if you are seriously suggesting that Crafty is 400 points weaker than
Rebel, but if you are, Crafty would be expected to score 9% on equal hardware.
This is insane.
> My suggestion is that even if Crafty wins one game in the time-odds
> match, but loses every game in the reference series, that there must be
> some advantage in the greater time allotment. Since the win would
> indicate an increase in the playing strength. The amount of the
> increase would be difficult to determine, however, if Crafty loses every
> game in the reference match because the "Elo" difference could be 400 or
> it could be 800, without wins, it is impossible to say.
The odds are almost zero that Crafty would lose every game in the reference
series, assuming that the reference series is of decent length.
When Crafty started playing on ICC it was pretty soft and squishy. It got a
reputation for not being very strong. This reputation has stuck with it in some
circles.
Said in caps so perhaps people will hear: Crafty is STRONG.
> It may be that Rebel will win the match, the important question to
> answer is: did Crafty's strength increase due to the increased amount of
> time it spent solving the problem? and if so, how much did it increase?
> as it is these questions which will answer the original controversy of
> whether or not searching deeper and deeper will generate a stronger
> program.
The biggest problem with this match is the difficulty of obtaining a reasonable
number of games at the long time control.
bruce
Richard A. Fowell
Peter Fendrich
brucemo wrote:
>
> Jack Nerad wrote:
> >
> > Robert Hyatt wrote:
> > >
> > > Jack Nerad (JNE...@concentric.net) wrote:
> > > : The point of the match is to see how many points stronger a program
> > > : becomes given greater amount of search time, correct? The series of
> > > : games played after the ten game time-odds match will be used as a
> > > : reference for judgement of this point. My question is this: will only
> > > : ten games in the time-odds match provide an accurate enough measure of
> > > : this statistically?
>
> No.
>
> If you run a 10-game match, assuming 40% drawing probability, and the opponents are
> exactly the same strength, the match score will be either 40%, 45%, 50%, 55%, or
> 60%, only about 70% of the time. This means that 30% of the time one of the
> programs will score 65% or better, which is at least 100 rating points indicated
> difference.
> The biggest problem with this match is the difficulty of obtaining a reasonable
> number of games at the long time control.
>
> bruce
I think it's possible to get some idea of the reliability of the match
result,
by studying each game. The thread about this will of cource be very
long...
Another possibility would be to check how often Crafty changes its move
from
3 minutes until the move is actually done and how much the eval has
changed.
This will give some hint of how much Crafty 'thinks' the gain of 3-4(?)
ply is.
Ed and Bob, do you log the full Crafty analysis?
Alltogether with the match result, games analysis and the Crafty log it
might
be possible to find some sort of pattern.
--
J-P Fendrich
Robert Hyatt
Peter Fendrich (j...@vd.volvo.se) wrote:
: I think it's possible to get some idea of the reliability of the match
: result,
: by studying each game. The thread about this will of cource be very
: long...
: Another possibility would be to check how often Crafty changes its move
: from
: 3 minutes until the move is actually done and how much the eval has
: changed.
: This will give some hint of how much Crafty 'thinks' the gain of 3-4(?)
: ply is.
: Ed and Bob, do you log the full Crafty analysis?
: Alltogether with the match result, games analysis and the Crafty log it
: might
: be possible to find some sort of pattern.
Yes it does.
I'm interested in three things that I see happen on occasion:
1. how often does the deeper search produce a better move, by spotting
a tactic, or by spotting a positional sequence that gains an advantage.
IE: metric 1: how often does the deeper search improve on what would
have been played by a shallower search.
2. how often does the deeper search produce a worse move, by spotting
a deep tactic or positional sequence that it discovers is unavoidable
(*if* the opponent can see it of course) and then changing its mind to
a simple move that gives away the material or advantage easily, rather
than making the opponent work for it. I've been burned here on many
occasions on ICC, at ACM events and at human chess events, because a
program, once it sees the roof fall in "way out there" assumes perfect
play the opponent. If one is way out-searching the other, the minimax
search is too pessimistic.
3. how many other cases are there. That is, where searching deeper just
burned a lot of time, and didn't change anything at all (same move, but
maybe a more accurate eval)...
after game 1 ends, I can post Crafty's log if everyone wants to see it, or
I can put it on the ftp machine (which might be a better way to conserve
usenet news bandwidth. In any case, I agree that careful analysis will be
interesting as it will likely be the first time we've seen a game played
exactly under these conditions...
Komputer Korner
brucemo wrote:
>
snipped
> > My suggestion is that even if Crafty wins one game in the time-odds
> > match, but loses every game in the reference series, that there must be
> > some advantage in the greater time allotment. Since the win would
> > indicate an increase in the playing strength. The amount of the
> > increase would be difficult to determine, however, if Crafty loses every
> > game in the reference match because the "Elo" difference could be 400 or
> > it could be 800, without wins, it is impossible to say.
>
> The odds are almost zero that Crafty would lose every game in the reference
> series, assuming that the reference series is of decent length.
>
> When Crafty started playing on ICC it was pretty soft and squishy. It got a
> reputation for not being very strong. This reputation has stuck with it in some
> circles.
>
> Said in caps so perhaps people will hear: Crafty is STRONG.
>
> > It may be that Rebel will win the match, the important question to
> > answer is: did Crafty's strength increase due to the increased amount of
> > time it spent solving the problem? and if so, how much did it increase?
> > as it is these questions which will answer the original controversy of
> > whether or not searching deeper and deeper will generate a stronger
> > program.
>
> number of games at the long time control.
>
> bruce
Obviously longer time control games are more important for determining
ultimate strength, but no one has come up with a formula to weight
this variable.
--
Komputer Korner
The inkompetent komputer.
Komputer Korner
Robert Hyatt wrote:
>
>snipped
> : Jack Nerad
> snipped
> It certainly might happen. Of course, it would obviously *not* speak very
> well of Crafty. Nor anyone else for that matter, since Crafty played several
> very good programs in Jakarta and won most of the games... But you have to
> go figure anyway. For example, Crafty has blown MacChess out in the
> Korrespondence Kup, yet Rebel/MacChess is an apparent draw. That alone
> leads me to believe that while Rebel might well be better than Crafty, it
> is not a huge difference...
The MacChess-Rebel game was an exchange SLAV where MacChess with white
seemed to go for the draw right from the start. At this level if
white is going to play for the draw in an Exchange SLAV, there is
little that the Black side can do about it. And of course single games
aren't worth very much even if they are Korrespondence games. As for the
ICS, blitz time controls are a different game of chess than 40/2 or
Korrespondence.
Rolf Tueschen
Finally I composed my first musical. I could combine my interests in
sports/chess / music and justice.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Here quoting of the initial birthing of the first act.
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>Not necessarily, because there are *two* variables in this equation.
>If the test were Crafty at depth=10 vs Crafty at depth=15, there would
>be only *one* variable, the depth. Or if we tested Rebel at depth=10
>vs Rebel at depth=15, there would be only one variable. Here there are
>two. It doesn't take much to make the outcome of Prog.a vs Prog.b change
>in a significant way.
ROTFL with double inversed pirouette.
>However, 5-5 would simply mean that Crafty's eval is too much worse than
>Ed's to make the result useful.
ROTFL with toolooped rittberger's safari on the right igel-ed
stonewall- break- dance. Ha ha ha ha , seeing the light, seeing the light, ha
ha ha ha ... (three times forte altissimo)
-----------------------
Note all the above is true and in full accordance to the law.
--- งง 267a ---
(source the BH Alabama museum of computer chess science for advanced beginners)
Rolf Tueschen
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>If the experiment doesn't end pretty
>quickly, it will span a time-frame that covers multiple rebel and Crafty
>revisions. If the split is 8-2, the extra depth will be pretty convincing.
>If it is 2-8, then it will say that Rebel's evel is much better than Crafty's.
>If it is anything like 5-5 it will be inconclusive. Game one is interesting.
>It was played with Crafty not having a book
snipped
Two objections if allowed.
A) The match should be playes with the actual versions of the two prgs.
Otherwise it could lead to special doping issues ...
B) In difference to the logic used in Alabama the world would appreciate a
score of 5-5 rightly as a *triumph* for Rebel. Also a 4-6 or 3-7 lost should be
sort of impressive sign for the strength of Rebel.
It's not like in Alabama where a factor 100 advantage of the whites is known
as/called racial balance ... :))
An expected win of Rebel would show the huge gap between the performance of a
commercial topper and the academical approach specialized for rapids against
humans on ICC.
-----------------
Note, all the above is absolutely true otherwise I would never have posted it.
Following the rule of usenet. >>Only post s.th. you're 100% sure of. And you'd
proven in longterm physical experimental settings at least above 800 degrees of
Celsius.<<
--- ACCTF §§ 267a of Bob Hyatt's Alabama Computer Chess Traffic Flaws ---
Howard Exner
Wouldn't having identical programs(like in twin studies) play this
time handicap match allow for greater predictability, since a
variable( two extremely different programs) is eliminated? This of
course would not be as much fun to watch.
Robert Hyatt
Howard Exner (hex...@dlcwest.com) wrote:
: Robert Hyatt wrote:
: >I now have literally
: > dozens of games against most of the current crop of programs,
: including
: > 40+ against Hiarcs 6. The surprising result? CM5000 is *still*
: the
: > strongest opponent (from crafty's perspective).
: What are the time controls of these games? Are there 40/2 games with
: Crafty vs. "current crop"? If so could you post these 40/2 games?
: Something I do for fun with the SSDF results is create my own
: database using just the Top 35 programs (ending with Fritz 3/dx-66 at
: #35). The results were usually within the (+ -)10 range with the
: exception of three programs:
: Latest March SSDF rating is
The games are a wild mix... from sort of longish blitz (5 5 ICC time
control) to 30 30, although some are as long as 60 60. I try to avoid
letting Crafty play anything that long as the "humans get restless" and
start leaving me messages about "why can't the comp vs comp games be
played offline so we can play Crafty?"
: CM5000 = 2386
: R30 v2.5 = 2365
: Fritz 4 = 2336
: Eliminating all opposition weaker than Fritz 3 / DX-66 from the pool
: results in:
: CM5000 = 2363
: R30 v2.5 =2401
: Fritz 4 = 2384
: Not sure why these three programs vary so much but it's interesting
: to see CM5000 dropping below it's older brother and Fritz 4. Most
: likely a larger pool of games would be more conclusive. The nice
: feature of the SSDF is there posting of individual match scores. Many
: programs have a difficult opponent just like yours against CM5000.
: From Genius 3's perspective MCPro 6 is the "strongest opponent",while
: from Genius 4's it's Rebel 6.0 that is the "strongest". Fortunately
: the pool is varied in the SSDF testing.
I've noticed this. That's why I've tried to be *very* careful to always
include the "from Crafty's perspective" as that is the best measure I
have. I remember some program Lonnie was using that Crafty was beating
pretty consistently and Ferret was having one hell of a time with it for
some odd reason. Bruce and I chatted about it once or twice, he might
remember which program it was.
I'd bet, if you look hard enough, that you can find the "impossible" in
comp vs comp games... A beats B consistently, B beats C consistently, and
C beats A consistently. Makes little sense, but who knows... I just try to
watch and fix what I can. :)
Robert Hyatt
Howard Exner (hex...@dlcwest.com) wrote:
: Wouldn't having identical programs(like in twin studies) play this
: time handicap match allow for greater predictability, since a
: variable( two extremely different programs) is eliminated? This of
: course would not be as much fun to watch.
Yes it would as I've mentioned. But this adds a little fun to the
mix. Boring as hell to watch A play A with a huge time handicap. This
way at least Ed and I will look at the games under a microscope later
to see what the extra depth did or didn't do. Others will be welcome
to do the same of course. We probably need to somehow capture the
iteration by iteration score/pv info for crafty and interlace it with
the same for rebel so it can be "read" (sort of..)
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
While I agree with your analysis, your inKompetence fell out with the last
sentence about ICS blitz time controls... what is that all about since I
didn't see anyone mention blitz time controls or ICC anywhere? :)
Howard Exner
Robert Hyatt wrote:
>I now have literally
> dozens of games against most of the current crop of programs,
including
> 40+ against Hiarcs 6. The surprising result? CM5000 is *still*
the
> strongest opponent (from crafty's perspective).
What are the time controls of these games? Are there 40/2 games with
Crafty vs. "current crop"? If so could you post these 40/2 games?
Something I do for fun with the SSDF results is create my own
database using just the Top 35 programs (ending with Fritz 3/dx-66 at
#35). The results were usually within the (+ -)10 range with the
exception of three programs:
Latest March SSDF rating is
CM5000 = 2386
Jesper Antonsson
In article <5gpqf9$4...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>2. how often does the deeper search produce a worse move, by spotting
>a deep tactic or positional sequence that it discovers is unavoidable
>(*if* the opponent can see it of course) and then changing its mind to
>a simple move that gives away the material or advantage easily, rather
>than making the opponent work for it. I've been burned here on many
>occasions on ICC, at ACM events and at human chess events, because a
>program, once it sees the roof fall in "way out there" assumes perfect
>play the opponent. If one is way out-searching the other, the minimax
>search is too pessimistic.
Has anyone experimented with remedies for this? An idea just struck me that
might help. If we define x as the number of plies before the last node (in the
path from root to current leaf) with a material balance equal to that of the
root. Then, if we at a leaf find that material is lost, we add say 0.05*x to
that position as a "distance to loss"-bonus. I chose 0.05 because with the
maximum search depth of todays pc-programs, the bonus wont ever be (much)
more than half a pawn.
I haven't got enough experience to know if this would be doable, or if
there are any fundamental flaws that I overlooked, but it seems quite
reasonable. Will there be a performance hit? Will it work?
If it does work, it should be good against slower programs and most
importantly against humans.
/Jesper Antonsson
Robert Hyatt
Jesper Antonsson (jes...@lysator.liu.se) wrote:
I'm not sure it's that easy. Here's something Berliner tried in HiTech and
which I tried in Cray Blitz:
suppose we are searching to depth 12. for plies 1-11 we liked move A, with
an eval of -.300, sort of bad, but not fatal. Then at ply 12, move A drops to
-1.400, and after searching move B, we improve this to -1.200... What Hans
proposed, and which I tried, was to reject B and stick with A. Because often,
A is the better move. Remember that an 11 ply search couldn't find anything
wrong with A, but a 12 ply search did. Therefore it is pretty deep. And
the 11 ply search didn't find anything good about B, which probably means it
saw that B lost a pawn quickly. So, which would you rather play? The move
that takes a *deep* search to see the loss, or something else? The HiTech
idea was to stick with A.
In the Rebel match this wouldn't work (it is not yet in Crafty anyway) because,
to follow the above analogy, when Crafty reaches Rebel's search depth with Move
A, and finds it is ok, but fails at depth 11 and wants to switch, we could prevent
it from doing so. But the search continues for 12, 13, 14, and 15 plies. And you
can't use that same reasoning now, because 10 looks like a "shallow search" when
you think about the current 15 ply search. The idea was a kludge that handled the
case of failing low at the last minute, and then finding another move that was
only slightly better, but was much more obvious to the opponent...
However, your idea is certainly worth testing some. I'll put it on my list.
What would be interesting to do is to add this, then play that version against
an older version, giving the new code a couple of extra plies of search, to
see if it would avoid screwing up when it sees something that the opponent can't
possibly find. Even better would be to figure out how deep the opponent is
searching (on average) and use that number. If your opponent is searching less
deeply than you are, it is possible, because if he drops a pawn, you can figure
out how deep you searched to find the win of the pawn, and know that he searched
at least one ply less than that or he would not have lost it. Interesting ideas
of course. The issue is "discernability" and it's both sticky and critical. Of
course, if you make no mistakes, then it probably won't be a problem. But if you
do, this might be one way to salvage a way out.
One interesting test I'm going to do after this first game is over, is to analyze
the game using Rebel and Crafty here, and see what would happen if Crafty didn't
change it's mind when it sees something bad at depth 15, and, instead, plays the
move it liked at 12-13. Should give some insight into whether this is worth
pursuing or not.
Dan Schmidt
I've been wondering if it's possible to play as trickily as possible
if you know that your opponent (human or computer) is not looking as
deep as you are.
What you want to do is induce the opponent to play a bad move on his
next move, moving into a subtree that is only good for you (and bad
for him) because of things at the "bottom" of it, that you see and he
doesn't.
(Hand-waving wildly) For example, you could search 12 ply deep, but
only do the "min" in the "minimax" down to 8 ply, to simulate that the
opponent is only doing evaluations to around 8 ply in the tree and
will likely miss things deeper?
Or can you give a little bonus to moves whose eval changes markedly
(for the better) once you get past ply 8?
Dan
Vincent Diepeveen
>I've been wondering if it's possible to play as trickily as possible
>if you know that your opponent (human or computer) is not looking as
>deep as you are.
If your program is running on 2 computers, one looking ahead x
ply, the other x + n ply, one could do this.
If 2 different programs are playing, no way you can do this:
take for example larsen test 99: Rxd3!
This is a move which wins material, as white has a mate thread.
Most programs need around 11 ply to find this. Suppose your looking
in a position x, where you want to play move m, which gets you to
position y which is the same as that larsen 99 position.
At 12 ply your program sees the problem.
As you search deeper than your opponent, you're playing rebel, which runs on a 486-4MB,
and you are running on a Cray, you estimate that it will not see it.
Now the problem: Rebel sees this trick already at 6 ply...
...poor you.
So the problem in this case is extensions, and evaluation.
For example: NOLOT position 10: Rxf7!!, is found by Diep at depth=10.
What depth does for example Cray Blitz need to find this? It's just
a matter of capturing a lot of pieces, few checks, an in between move,
and another few captures.
I'm sure that currently their is no hardware used by any program
which can gamble that the selective search of the opponent will not see
it. These selective searches usually can make up few plies.
You don't outsearch an opponent deeper than few plies.
Take for example my draughtsprogram. It has been the case that we
outsearched opponents by 6 plies (we played on a Pro, most played on a
pentium, some on a 486). We did not win any game tactically
last blitz championship (rapid would be a better word: about 25 minutes a game).
It appeared however that this
search-advantage was only useful in the endgame,
where no of the 2 programs had much factors in their evaluation.
In the endgame we drew a lost position by simply outsearching the opponent, and
we won an equal endgame.
The lack of a good mobility in our draughtsprogram seemed to be the problem
in the opening/middlegame. We simply do not use it! The stupid program simply wanted
to advance all pieces, so the only game we lost, we lost because the program
played bad in opening. Then we came into a complex (still objectively bad) middlegame
where the program outsearched and outsearched. Didn't work. Somehow it seemed
that all these extra plies of us didn't give much advantage. It just PREVENTED
a quick loss.
So if a certain program searches far deeper than the competition, i wouldn't say
that it could be able to take advantage of it. It just can prevent some bad things
with it!
>What you want to do is induce the opponent to play a bad move on his
>next move, moving into a subtree that is only good for you (and bad
>for him) because of things at the "bottom" of it, that you see and he
>doesn't.
>
>(Hand-waving wildly) For example, you could search 12 ply deep, but
>only do the "min" in the "minimax" down to 8 ply, to simulate that the
>opponent is only doing evaluations to around 8 ply in the tree and
>will likely miss things deeper?
>
>Or can you give a little bonus to moves whose eval changes markedly
>(for the better) once you get past ply 8?
>
>Dan
--
+----------------------------------------------------+
| Vincent Diepeveen email: vdie...@cs.ruu.nl |
| http://www.students.cs.ruu.nl/~vdiepeve/ |
+----------------------------------------------------+
Robert Hyatt
Dan Schmidt (df...@harmonixmusic.com) wrote:
: I've been wondering if it's possible to play as trickily as possible
: if you know that your opponent (human or computer) is not looking as
: deep as you are.
: What you want to do is induce the opponent to play a bad move on his
: next move, moving into a subtree that is only good for you (and bad
: for him) because of things at the "bottom" of it, that you see and he
: doesn't.
: (Hand-waving wildly) For example, you could search 12 ply deep, but
: only do the "min" in the "minimax" down to 8 ply, to simulate that the
: opponent is only doing evaluations to around 8 ply in the tree and
: will likely miss things deeper?
: Or can you give a little bonus to moves whose eval changes markedly
: (for the better) once you get past ply 8?
: Dan
This needs thought, because it is also similar to the well-known database
(endgame) problem where assuming the opponent has one is a mistake. Making
the move leading to the longest mate might also be making the move that leads
to the most obvious way for the opponent to see how to win...
if you have perfect knowledge, and assume your opponent does too, then things
work well as they are. But if you know your opponent has less, then the current
minimax strategies are not as good...
Robert Hyatt
Vincent Diepeveen (vdie...@cs.ruu.nl) wrote:
: >I've been wondering if it's possible to play as trickily as possible
: >if you know that your opponent (human or computer) is not looking as
: >deep as you are.
: If your program is running on 2 computers, one looking ahead x
: ply, the other x + n ply, one could do this.
: If 2 different programs are playing, no way you can do this:
: take for example larsen test 99: Rxd3!
: This is a move which wins material, as white has a mate thread.
: Most programs need around 11 ply to find this. Suppose your looking
: in a position x, where you want to play move m, which gets you to
: position y which is the same as that larsen 99 position.
: At 12 ply your program sees the problem.
: As you search deeper than your opponent, you're playing rebel, which runs on a 486-4MB,
: and you are running on a Cray, you estimate that it will not see it.
: Now the problem: Rebel sees this trick already at 6 ply...
: ...poor you.
: So the problem in this case is extensions, and evaluation.
: For example: NOLOT position 10: Rxf7!!, is found by Diep at depth=10.
: What depth does for example Cray Blitz need to find this? It's just
: a matter of capturing a lot of pieces, few checks, an in between move,
: and another few captures.
depth=8, <1 second. On a C90. However, this is an easy problem in this
set, and also fits one extension we did in CB for many years, that of treating
any capture next to the king as a check, which means the initial move Rxf7
causes an extension immediately. I'm running this on Crafty right now, but
it didn't find it at depth=11. However I'm a lot more conservative in Crafty
with how I extend since I don't have anywhere near the horsepower I did with
the Cray... just finished depth=12 and also didn't find it... probably a result
of null-move, razoring, and other things added together... I'll have to check to
see what is blocking the solution from Crafty's sight.
: I'm sure that currently their is no hardware used by any program
: which can gamble that the selective search of the opponent will not see
: it. These selective searches usually can make up few plies.
: You don't outsearch an opponent deeper than few plies.
Unless you are Deep Blue of course...
: Take for example my draughtsprogram. It has been the case that we
: outsearched opponents by 6 plies (we played on a Pro, most played on a
: pentium, some on a 486). We did not win any game tactically
: last blitz championship (rapid would be a better word: about 25 minutes a game).
: It appeared however that this
: search-advantage was only useful in the endgame,
: where no of the 2 programs had much factors in their evaluation.
: In the endgame we drew a lost position by simply outsearching the opponent, and
: we won an equal endgame.
: The lack of a good mobility in our draughtsprogram seemed to be the problem
: in the opening/middlegame. We simply do not use it! The stupid program simply wanted
: to advance all pieces, so the only game we lost, we lost because the program
: played bad in opening. Then we came into a complex (still objectively bad) middlegame
: where the program outsearched and outsearched. Didn't work. Somehow it seemed
: that all these extra plies of us didn't give much advantage. It just PREVENTED
: a quick loss.
: So if a certain program searches far deeper than the competition, i wouldn't say
: that it could be able to take advantage of it. It just can prevent some bad things
: with it!
: >What you want to do is induce the opponent to play a bad move on his
brucemo
Dan Schmidt wrote:
>
> I've been wondering if it's possible to play as trickily as possible
> if you know that your opponent (human or computer) is not looking as
> deep as you are.
>
> next move, moving into a subtree that is only good for you (and bad
> for him) because of things at the "bottom" of it, that you see and he
> doesn't.
>
> (Hand-waving wildly) For example, you could search 12 ply deep, but
> only do the "min" in the "minimax" down to 8 ply, to simulate that the
> opponent is only doing evaluations to around 8 ply in the tree and
> will likely miss things deeper?
>
> Or can you give a little bonus to moves whose eval changes markedly
> (for the better) once you get past ply 8?
Interesting idea but you'd lose a lot of efficiency. For every move you
are considering making, you'd have to get a nearly exact value for each
of the possible replies, and detect conditions where the opponent has a
reply that's hard to see that brings the score slightly above the best
reply to the move that you thing is really objectively best, and at
least one other move that is easy to see, whose score drops a lot in
later iterations.
This degree of exactness might be expensive.
I realized that my first paragraph is hard to understand, so I'll give
an example, by attributing silly scores to some real moves. Let's say
you're planning on playing 1. e4, and you have this scored as +0.20.
Let's say that you notice that if you play 1. d4, 1. ... d5 looks like a
great reply until depth 12 (scoring +0.05 from your point of view), at
which point the score increases (from your point of view to +0.50).
But, at that point, the reply 1. ... Nf6 is observed to result in a
score of +0.15 from your point of view.
So, if you play 1. e4 you'll probably be ahead by +0.20, but if you play
1. d4, and your opponent searches as deeply as you did (actually a ply
shallower, since they don't have to search your move, you've already
made it on the board), you'll end up facing 1. ... Nf6, and will be in a
slightly weaker but still playable position. But if you can be sure you
out-search them, you'll probably get 1. ... d5, which leaves you in
great shape.
Sounds complicated. I wonder if there are many practical cases?
bruce
Andreas Mader
Jack Nerad wrote:
>
> The point of the match is to see how many points stronger a program
> becomes given greater amount of search time, correct? The series of
> games played after the ten game time-odds match will be used as a
> reference for judgement of this point. My question is this: will only
> ten games in the time-odds match provide an accurate enough measure of
> this statistically?
>
If you only take the "pure" result, there are not enough games to prove
anything, although some people will try to do so... :-(
But when you look at the moves and the "critical positions", analyse the
reasons of a win / loss and so on you will probably get more insights.
Best wishes
Andreas Mader
eric...@sprynet.com
Robert Hyatt wrote:
>
> You can run this test yourself if you have Crafty and xboard/winboard...
>
> try
>
> xboard -mm -fcp "crafty xboard st=30" -scp "crafty xboard st=300" and see
> what happens. 10:1 time odds. I do this a lot, and the 10-1 odds is an
> overwhelming advantage...
It seems to me that if the question is simply "at higher plies(the
number generally looked at by commercial programs or crafty on a PC)
how much more would an extra 3 or 4 plies be worth in terms of
strength?" then an experiment like the one you speak of, a match
between the same program with time differences to generate the stated
ply difference, would be a better way to find an answer. In the case
of Crafty-Rebel, the evals of the 2 programs are going to have to be
taken into consideration regardless of the outcome of the "control"
match at equal time controls. I assumed that the debate about "DB vs.
the micros" was one that concentrated on the hardware advantage,
without calling into question the software.
Simon Read
t...@mail.htp.com (Tim Mirabile) wrote:
> [Fischer-Reshevsky USA championship 1962]
>We have Fischer's very detailed notes to go by
Is this game in "My 60 memorable games" ?
If not, where do Fischer's notes appear?
Simon
dong...@aol.com
> In any case, I agree that careful analysis will be
>interesting as it will likely be the first time we've seen a game played
>exactly under these conditions...
>
>
This match up is quite analogous to saying what would happen if you could pit 2 predators of different niches against each other? E.g. Tiger vs Shark.
A comparison I once took notice of in a past issue of "Chess Life", when the topic of who was stronger, a correspondence world champion, or a OTB world champion? The suggestion was that with the advent of chess computers we could for the first time in history make an attempt at finding out? Cool.
yours in chess,
Don
Ramsey MN USA
Harald Faber
quoting a mail from hyatt # crafty.cis.uab.edu
Hello Robert,
RH> From: hy...@crafty.cis.uab.edu (Robert Hyatt)
RH> Subject: Re: The Crafty-Rebel NPS challenge started!!!
RH> Organization: CIS, University of Alabama at Birmingham
RH> The games are a wild mix... from sort of longish blitz (5 5 ICC time
RH> control) to 30 30, although some are as long as 60 60. I try to avoid
RH> letting Crafty play anything that long as the "humans get restless" and
RH> start leaving me messages about "why can't the comp vs comp games be
RH> played offline so we can play Crafty?"
I beg your pardon but I didn't find out the meaning of the time controls
like 5 5, 30 30 etc. What do they stand for?
Harald
e-mail:_ Harald...@t-online.de
--
Harald Faber
quoting a mail from blah # blurgh.bleah.retch
Hello Simon,
SR> From: Simon Read <bl...@blurgh.bleah.retch>
SR> Subject: Re: The Crafty-Rebel NPS challenge started!!!
SR> Organization: Serious Cybernetics Corporation
SR> t...@mail.htp.com (Tim Mirabile) wrote:
SR> > [Fischer-Reshevsky USA championship 1962]
SR> >We have Fischer's very detailed notes to go by
SR> Is this game in "My 60 memorable games" ?
Yes it is.
SR> Simon
Robert Hyatt
Harald Faber (Harald...@p21.f2.n1.z1001.fidonet.org) wrote:
: quoting a mail from hyatt # crafty.cis.uab.edu
: Hello Robert,
: RH> From: hy...@crafty.cis.uab.edu (Robert Hyatt)
: RH> Subject: Re: The Crafty-Rebel NPS challenge started!!!
: RH> Organization: CIS, University of Alabama at Birmingham
: RH> The games are a wild mix... from sort of longish blitz (5 5 ICC time
: RH> control) to 30 30, although some are as long as 60 60. I try to avoid
: RH> letting Crafty play anything that long as the "humans get restless" and
: RH> start leaving me messages about "why can't the comp vs comp games be
: RH> played offline so we can play Crafty?"
: I beg your pardon but I didn't find out the meaning of the time controls
: like 5 5, 30 30 etc. What do they stand for?
sorry... they are "ICC/FICS" terminology. the first parameter is the
initial clock time in minutes. the second is the time per move added to
the clock after each move is made.
for a 30 30 game, the clock starts at 30 minutes, and decrements second by
second while I think. After I make a move, 30 seconds are added back to my
clock. Ditto for the opponent. A 30 30 game can take quite a while, but
the main point is they are much different from traditional blitz chess, in
that you don't have to run out of time, ever, because you always have 5
seconds (or whatever the increment is) per move added to your clock.