Rebel-Crafty NPS Challenge

32 views
Skip to first unread message

Amir Ban

unread,
Mar 19, 1997, 3:00:00 AM3/19/97
to

Congratulations Ed.

The first game is really proving your point. Whom did you bribe to make this happen ? I
knew that a 100-time speedup can mean nothing against a master, but I was skeptical
about computer-computer games.

It's obvious that Crafty doesn't understand this game, and this is making all the
difference so far: Refusing to trade queens, that horrible 22... Rb8 (how about Rc8 and
putting some pressure on the C file weakness ?). Bob posted here that Crafty counts the
passed e5 pawn as a big asset. It's really killing Black's game and Black should get
rid of it when given the chance. The chance, I think, came and went with 25. h5 gxh5?
missing 25... e4 to create an unclear mess. Anyone has a deep analysis of this ?

I wonder how many times in this match Rebel would be allowed to play such a game.

Another point for the NPS debate, and the importance of evaluations: A human master as
Black here would be thinking how he can offer a pawn to free himself. If Black can
stabilize his game with a pawn less, he has chances. Rebel with current eval (0.81)
would maybe take it, maybe not. Crafty with an eval of 0.46 would grab it, and
certainly doesn't think of offering it.

Amir

Robert Hyatt

unread,
Mar 19, 1997, 3:00:00 AM3/19/97
to

Amir Ban (ami...@msys.co.il) wrote:
: Congratulations Ed.

: The first game is really proving your point. Whom did you bribe to make this happen ? I
: knew that a 100-time speedup can mean nothing against a master, but I was skeptical
: about computer-computer games.

: It's obvious that Crafty doesn't understand this game, and this is making all the
: difference so far: Refusing to trade queens, that horrible 22... Rb8

Rb8's not so terribly horrible. If you search to about ply=16, I believe
that Rc8 loses something. We ran a lot of analysis here one night, because
I liked Rc8 (as did crafty until some ungodly deep depth.)

However, as I posted before, I'd bet that if you let any program play the
black side of this, from about move 20 or so, they'd lose pretty quickly
because some of the tactics are quite deep. I've tried it with genius and
it simply doesn't find many of the right moves for black unless you give it
the same huge time handicap. Point being that the deeper search is helping,
but in this game (at least) Crafty was more concerned about the e-pawn than
the kingside attack. Game's not over by any stretch however...

One minor point here is that this game is not a convincing argument that
5 plies means nothing. In this game, Crafty happens to value passed pawns
quite highly, a result of tuning to play against humans. As a result, if
you count pawns, you'll notice who has one and who doesn't. Now whether
or not that is the goal in *this* game is obviously another issue. But so
far, Crafty has been able to accomplish what it "thought" was the right
things. If it's eval is out of whack, that's another issue altogether.
However if you want to join the bandwagon that 100X speed means nothing,
you're free to do so. But I'll bet you are just like me, when the p6/400
comes out, you'll be right there on it as well, to get that paltry 2x
speedup that means "less than nothing." :)

As far as trading queens, Crafty generally won't do that unless one side's
king is significantly more exposed than the other's. In this game, both
kings are somewhat exposed. It's playing just like I'd want to see it play
against a human. Keep it complicated and wait for a mistake. Against a
computer that might not be the right strategy. Against humans it is working
very well.


: (how about Rc8 and

: putting some pressure on the C file weakness ?). Bob posted here that Crafty counts the
: passed e5 pawn as a big asset. It's really killing Black's game and Black should get
: rid of it when given the chance. The chance, I think, came and went with 25. h5 gxh5?
: missing 25... e4 to create an unclear mess. Anyone has a deep analysis of this ?

: I wonder how many times in this match Rebel would be allowed to play such a game.

: Another point for the NPS debate, and the importance of evaluations: A human master as
: Black here would be thinking how he can offer a pawn to free himself. If Black can
: stabilize his game with a pawn less, he has chances. Rebel with current eval (0.81)
: would maybe take it, maybe not. Crafty with an eval of 0.46 would grab it, and
: certainly doesn't think of offering it.

Current analysis by Crafty is that it seems to be planning on sacrificing the
exchange to rip the bishop attacking f7, and picking up the c2 pawn with check
as well, although I'm not going to let it run for 10 hours here to produce
exactly what Ed's seeing over there. Whether it should give up the e-pawn
instead is a question of course. In Crafty's case, it's holding on to that
for dear life...


: Amir

Ed Schroder

unread,
Mar 20, 1997, 3:00:00 AM3/20/97
to

From: Amir Ban <ami...@msys.co.il>

: Congratulations Ed.

: The first game is really proving your point. Whom did you bribe to
: make this happen ? I knew that a 100-time speedup can mean nothing
: against a master, but I was skeptical about computer-computer games.

: It's obvious that Crafty doesn't understand this game, and this is
: making all the difference so far: Refusing to trade queens, that

: horrible 22... Rb8 (how about Rc8 and putting some pressure on the C

: file weakness ?). Bob posted here that Crafty counts the passed e5
: pawn as a big asset. It's really killing Black's game and Black should
: get rid of it when given the chance.

: The chance, I think, came and went with 25. h5 gxh5? missing 25... e4
: to create an unclear mess. Anyone has a deep analysis of this ?

Rebel8 expected (would have played) this move with incredible wild
tactics.

: I wonder how many times in this match Rebel would be allowed to play
: such a game.

: Another point for the NPS debate, and the importance of evaluations:
: A human master as Black here would be thinking how he can offer a pawn
: to free himself. If Black can stabilize his game with a pawn less, he
: has chances. Rebel with current eval (0.81) would maybe take it, maybe
: not. Crafty with an eval of 0.46 would grab it, and certainly doesn't
: think of offering it.

: Amir

At the moment game-1 looks as follows:

[Event "The Crafty-Rebel NPS challenge"]
[Site "Deventer, The Netherlans"]
[Date "1997.03.07"]
[Round "1"]
[White "REBEL 8.0"]
[Black "CRAFTY 11.17"]
[Result "*"]
[ECO "B20"]

1. e4 c5 2. b3 d6 3. Bb2 Nf6 4. Nc3 g6 5. g3 Bg7 6. Bg2 0-0 7. Nge2 e5
8. 0-0 Nc6 9. f4 c4 10. Kh1 cxb3 11. axb3 Bg4 12. h3 Bxe2 13. Nxe2 Re8
14. f5 d5 15. exd5 Nxd5 16. Kh2 a6 17. Be4 Qd7 18. fxg6 hxg6 19. Nc3 Nxc3
20. dxc3 Qc7 21. Bd5 Nd8 22. Qf3 Rb8 23. Rad1 Ne6 24. h4 b5 25. h5 gxh5
26. Qxh5 Rb6 27. Kg2 Nd8 28. Bc1 a5 29. Be4 Ne6 30. Be3 *

Rebel8 score 1.41
Crafty after 7 hours is on ply 14, score -1.547 planning Ra6 failing low.

One thing game-1 already proves...
In this game Crafty thinks average 4 plies deeper than Rebel due to the
100 x time advantage. Still Crafty is in deep trouble. Bob said Rebel's
evaluation function is better than Crafty's evaluation function. I agree
with Bob for game-1. Game2-10 will tell us more.

Coming to my point: Doesn't game-1 already prove that an evaluation
function is (can be) worth many many plies?

I am not speaking about Rebel but generally...

- Ed Schroder -

Rebel's chances to win this game are looking good but I think
Crafty can hold the king attack and perhaps force an ending with one
(or two?) pawns down. Game-1 is certainly not over yet.

Robert Hyatt

unread,
Mar 20, 1997, 3:00:00 AM3/20/97
to

Ed Schroder (rebc...@xs4all.nl) wrote:

: At the moment game-1 looks as follows:

: [Event "The Crafty-Rebel NPS challenge"]
: [Site "Deventer, The Netherlans"]
: [Date "1997.03.07"]
: [Round "1"]
: [White "REBEL 8.0"]
: [Black "CRAFTY 11.17"]
: [Result "*"]
: [ECO "B20"]

: 1. e4 c5 2. b3 d6 3. Bb2 Nf6 4. Nc3 g6 5. g3 Bg7 6. Bg2 0-0 7. Nge2 e5
: 8. 0-0 Nc6 9. f4 c4 10. Kh1 cxb3 11. axb3 Bg4 12. h3 Bxe2 13. Nxe2 Re8
: 14. f5 d5 15. exd5 Nxd5 16. Kh2 a6 17. Be4 Qd7 18. fxg6 hxg6 19. Nc3 Nxc3
: 20. dxc3 Qc7 21. Bd5 Nd8 22. Qf3 Rb8 23. Rad1 Ne6 24. h4 b5 25. h5 gxh5
: 26. Qxh5 Rb6 27. Kg2 Nd8 28. Bc1 a5 29. Be4 Ne6 30. Be3 *

: Rebel8 score 1.41
: Crafty after 7 hours is on ply 14, score -1.547 planning Ra6 failing low.

: One thing game-1 already proves...
: In this game Crafty thinks average 4 plies deeper than Rebel due to the
: 100 x time advantage. Still Crafty is in deep trouble. Bob said Rebel's
: evaluation function is better than Crafty's evaluation function. I agree
: with Bob for game-1. Game2-10 will tell us more.

: Coming to my point: Doesn't game-1 already prove that an evaluation
: function is (can be) worth many many plies?

Yes... and don't forget that I've never said otherwise. If you look at my
original statement "two equivalent programs... a 3-4-5 ply search advantage
by one is *very* significant."

The thing I see, which was a minor irritant in 11.17, is that due to a change
I made, the evaluation for passed pawns went somewhat higher than I planned. If
you look at each of the moves Crafty has made so far, they all revolve around the
passed e pawn. At nearly every move, as it searches deeper than Rebel by a big
margin, it's found a way to hold that pawn, even if it means giving up something
somewhere else. And it continued to do that until it was too late to save it,

I have fixed a couple of bugs that have caused a few problems in game 1, and will
have an executable ready for when game 2 begins. The main one is that if the
game has to be re-started, Crafty won't spend 2x the normal time on the first
move after the restart, which is a big waste here since the reason for doing this
is to offset the fact that the trans/ref table has no useful info in it. After
a huge search it *still* has no useful info in it. :) The next "problem" I
found is that when game 2 starts, with Crafty using a book (I hope it seems
to work this time) it can easily reach move 30 in book. That's going to make
the last 10 moves in the time control take *forever*. I had an upper bound on
the time for a single move of 2x the simple time-control average. That was
taken out before Jakarta, but is probably a good idea here, otherwise the first
non-book move could take 18000/10*1.4*2 minutes, which is going to be about
3 days or so...

: I am not speaking about Rebel but generally...

Hans-Henrik Grand

unread,
Mar 20, 1997, 3:00:00 AM3/20/97
to

>From: Amir Ban <ami...@msys.co.il>

>: Congratulations Ed.

>: Amir

>At the moment game-1 looks as follows:

>[Event "The Crafty-Rebel NPS challenge"]
>[Site "Deventer, The Netherlans"]
>[Date "1997.03.07"]
>[Round "1"]
>[White "REBEL 8.0"]
>[Black "CRAFTY 11.17"]
>[Result "*"]
>[ECO "B20"]

>1. e4 c5 2. b3 d6 3. Bb2 Nf6 4. Nc3 g6 5. g3 Bg7 6. Bg2 0-0 7. Nge2 e5
>8. 0-0 Nc6 9. f4 c4 10. Kh1 cxb3 11. axb3 Bg4 12. h3 Bxe2 13. Nxe2 Re8
>14. f5 d5 15. exd5 Nxd5 16. Kh2 a6 17. Be4 Qd7 18. fxg6 hxg6 19. Nc3 Nxc3
>20. dxc3 Qc7 21. Bd5 Nd8 22. Qf3 Rb8 23. Rad1 Ne6 24. h4 b5 25. h5 gxh5
>26. Qxh5 Rb6 27. Kg2 Nd8 28. Bc1 a5 29. Be4 Ne6 30. Be3 *

>Rebel8 score 1.41
>Crafty after 7 hours is on ply 14, score -1.547 planning Ra6 failing low.

At least they seem to agree abaut the eval now.
Previously crafty and Rebel disagreed a lot abaut the eval.

>One thing game-1 already proves...
>In this game Crafty thinks average 4 plies deeper than Rebel due to the
>100 x time advantage. Still Crafty is in deep trouble. Bob said Rebel's
>evaluation function is better than Crafty's evaluation function. I agree
>with Bob for game-1. Game2-10 will tell us more.

And given the result of the match at normal tournament level (4-0 in
Rebels favour), it proves that crafty is not tuned for playing an opponent
like Rebel (or Rebel simply is better).

>Coming to my point: Doesn't game-1 already prove that an evaluation
>function is (can be) worth many many plies?

So far it does. But I think Robert H. is still working on the eval that
has contained known bugs in the past. E.G. Crafty brought it's queen to
early in play which gave it serious problems in many games, when it came
early out of book. This proves that programmers should worry, abaut how
to make the eval good already, when making the search engine. Probably,
R. Hyatt has been doing this, but when reading the ChangeLog in the
main.c it seems, that speed is what the author considered most important
for the beginning.



>I am not speaking about Rebel but generally...

I think we all new that. But now it has been made even more clear.

>Rebel's chances to win this game are looking good but I think
>Crafty can hold the king attack and perhaps force an ending with one
>(or two?) pawns down. Game-1 is certainly not over yet.

But Rebel is still playing for the win, and is very unlikely to loose,
even when playing with 4 plies less than Crafty. Also Rebel is unlikely
to make a tactical blunder, as the search for Rebel already is deep.

Would Kasparov be able to win this position against Deep Blue ?
I think the Deep Blue team must be working harder on their software
rather than their hardware...

Perhaps the eval should be modified so that when playing a human opponent,
It would give at least a pawn to stear towards tactical complications.
Another good idea, suggested earlier here by Palle Mathiesen, would be
to exchange pawns in order to create open positions with more tactics to
follow.

Greetings /hhg
--
-=((--------------------------------------------------))=-
-==(((Your computer is the best. However, mine is better)))==-
-=((--------------------------------------------------))=-

Hans-Henrik Grand

unread,
Mar 20, 1997, 3:00:00 AM3/20/97
to

In <5grh51$n...@juniper.cis.uab.edu> hy...@crafty.cis.uab.edu (Robert Hyatt) writes:

>Hans-Henrik Grand (h...@daimi.aau.dk) wrote:
[snip]

>: Perhaps the eval should be modified so that when playing a human opponent,


>: It would give at least a pawn to stear towards tactical complications.
>: Another good idea, suggested earlier here by Palle Mathiesen, would be
>: to exchange pawns in order to create open positions with more tactics to
>: follow.

>Depends. Crafty does exactly this all the time. And generally won't try to
>trade queens unless it is in real trouble, which is the right strategy against
>a tactically weaker player (human). But against computers, it often does backfire.
>Chris's sort of search/eval plays right into Crafty's strength and the search speed
>of Crafty overwhelms it generally. Crafty's speculative eval plays right into
>Ferret's strength, and the same happens to Crafty. I carefully analyzed a set of
>4 games from last week, where crafty won 1 and lost 3 vs Ferret. In every game,
>Crafty was at +1 positionally (1 whole pawn), and in looking at the positions, I
>was not uncomfortable with any of them. Unfortunately, in three of them it
>ended up at +.3, which is not bad, until you count pawns and get -1... so it
>gave up a pawn in three games to open things up or create complications. And it
>got a typical "computer response" for doing so. I don't plan (at present) on
>trying to change the way I do business here, because I'm not nearly as interested
>in tuning for computers as I am in tuning for humans. And it's going to cause
>problems at times. The Shredder game at the last WMCCC event was a good example
>of how this can backfire, as crafty defended very well against a good attack, but
>then went for a complication of its own doing, after the attack was over, and
>found a way to ignore a fairly deep mate threat by playing too aggressively in the
>position. It's a problem. I'd even considered two evals, referenced by a pointer,
>so that it can be changed depending on whether Crafty plays a computer or not, but
>it somehow felt unsatisfying to do so.. :)
I really can't see anything wrong with that.
Playing computers is not the same as playing humans, and I play different
depending on the opponent. Having different evals seems like a good idea.
You might even consider having another eval for lightning (icc bullet),
and smaller hash tables, so the initialisation becomes faster.

Robert Hyatt

unread,
Mar 20, 1997, 3:00:00 AM3/20/97
to

Hans-Henrik Grand (h...@daimi.aau.dk) wrote:

: >From: Amir Ban <ami...@msys.co.il>

: >: Congratulations Ed.

: >: Amir

:

I also need to find out from Moritz what kind of book he sent to ed, and
whether or not it had the "learn.dat" imported into it. The book selection
algorithm depends heavily on the learned results. If there are none, then
it will play lots of bad openings until it learns what is what. a result
of 4-0 is unexpected, but not improbable. It will be more interesting if it
ends up 20-0. I have (so far) played 4 40/120 games with rebel 8 here on
equal hardware (P6/200) and the result was 1 win for Crafty, one draw, and
2 losses. They are difficult to play in manual mode of course...


: >Coming to my point: Doesn't game-1 already prove that an evaluation


: >function is (can be) worth many many plies?

: So far it does. But I think Robert H. is still working on the eval that
: has contained known bugs in the past. E.G. Crafty brought it's queen to
: early in play which gave it serious problems in many games, when it came
: early out of book. This proves that programmers should worry, abaut how
: to make the eval good already, when making the search engine. Probably,
: R. Hyatt has been doing this, but when reading the ChangeLog in the
: main.c it seems, that speed is what the author considered most important
: for the beginning.
:
: >I am not speaking about Rebel but generally...

: I think we all new that. But now it has been made even more clear.

: >Rebel's chances to win this game are looking good but I think
: >Crafty can hold the king attack and perhaps force an ending with one
: >(or two?) pawns down. Game-1 is certainly not over yet.

: But Rebel is still playing for the win, and is very unlikely to loose,
: even when playing with 4 plies less than Crafty. Also Rebel is unlikely
: to make a tactical blunder, as the search for Rebel already is deep.

However, the "unlikely to make a blunder" is probably an overstatement.
The easy way to confirm this is to run any program against the Nolot test
suite. and get the results for 10 plies, then for 11, then for 12, and
you'll see that the results improve with depth. And not making one of
those correct moves is a blunder. I don't buy the "tactical sufficiency"
idea at all. I agree that 1 more ply doesn't gain a whole lot. But two
is better, and 4-5 is much better tactically.


: Would Kasparov be able to win this position against Deep Blue ?


: I think the Deep Blue team must be working harder on their software
: rather than their hardware...

I don't think they've worked on the "hardware" since last match. The chip
was "done". But the hardware processor does have a "firmware" sort of evaluation,
and the SP's search code is written in C, so I'd suspect that all they've been
doing (all? seems to make the work sound trivial, but it is far from that) is
tuning the eval and working on the search extension code, and tuning the parallel
search better.

Moritz Berger

unread,
Mar 20, 1997, 3:00:00 AM3/20/97
to

On 20 Mar 1997 14:21:21 GMT, hy...@crafty.cis.uab.edu (Robert Hyatt)
wrote:
(posted and mailed)

>I also need to find out from Moritz what kind of book he sent to ed, and
>whether or not it had the "learn.dat" imported into it.

On the 2nd CD, I built a book with Crafty 11.17 under DOS. Since I
took it from my Crafty directory, I guess that it had the latest
learn.dat included, of course the most recent learn.dat should be
imported before starting the next game ...

> The book selection
>algorithm depends heavily on the learned results. If there are none, then
>it will play lots of bad openings until it learns what is what. a result
>of 4-0 is unexpected, but not improbable.

There are still many lemons in the huge Crafty book which show up
against opponents it didn't play previously on ICC (such as recent
beta versions of some other programs/books).

> It will be more interesting if it
>ends up 20-0. I have (so far) played 4 40/120 games with rebel 8 here on
>equal hardware (P6/200) and the result was 1 win for Crafty, one draw, and
>2 losses. They are difficult to play in manual mode of course...

I think that Crafty should be able to score 35% against Rebel, which
is nothing like a 20-0 result (indicating a 400+ ELO difference ...).

-------------
Moritz...@msn.com

Andrew John Walker

unread,
Mar 21, 1997, 3:00:00 AM3/21/97
to

Congratulations, the match web site just passed 100000 visits!
Watching that one tick over was rivetting :-)

--
* Andrew Walker *
* Department Of Physics * e-mail -- aj...@uow.edu.au
* Wollongong University *
* Australia *

brucemo

unread,
Mar 21, 1997, 3:00:00 AM3/21/97
to

Robert Hyatt wrote:

> I have fixed a couple of bugs that have caused a few problems in game 1, and will
> have an executable ready for when game 2 begins.

You might want to go with asymmetrical king safety, too. Better to play in the
center and out-search your opponent than it is to allow a lot of king-side
counterplay potentially.

bruce

Patrick Jagstaidt

unread,
Mar 22, 1997, 3:00:00 AM3/22/97
to Robert Hyatt

Robert Hyatt wrote:
[stuff skipped]

> I have fixed a couple of bugs that have caused a few problems in game 1, and will
> have an executable ready for when game 2 begins.
[more stuff skipped]

If you allow for bug fixes between games the whole match becomes a
moving target. It was already difficult to discriminate what was due to

1. Time odd
2. Difference in evaluation function efficacy

but now in between bug fixes are furthermore blurring the issue...

PS Is Ed allowed to enhance his program during the competition ?

If yes I assume that the version playing is not the regular (commercial)
Rebel 8 which is frozen but a sequel, maybe the "enhanced" Rebel 8 under
beta test on the server...

Best regards,
P. Jagstaidt

Robert Hyatt

unread,
Mar 22, 1997, 3:00:00 AM3/22/97
to

Patrick Jagstaidt (jags...@iprolink.ch) wrote:

: Robert Hyatt wrote:
: [stuff skipped]
: > I have fixed a couple of bugs that have caused a few problems in game 1, and will
: > have an executable ready for when game 2 begins.
: [more stuff skipped]

: If you allow for bug fixes between games the whole match becomes a
: moving target. It was already difficult to discriminate what was due to

: 1. Time odd
: 2. Difference in evaluation function efficacy

: but now in between bug fixes are furthermore blurring the issue...

Without the bug fixes, it is much worse. We are working on two problems
that I'm sending ed fixes on from time to time.

1. auto232 is still not right. And I've tried various things to see if it is
a timing-related problem, which it seems to be. The latest changes didn't help.

2. a fix to the "AN" command which auto232 uses to start a new game. It was
not resetting the time control correctly (I have never tested this since xboard
kills the old crafty and starts a new one for each and every game played on a
server.) As a result, crafty was getting lost in how to time each move. This
had no effect on the long game, but was causing problems in the calibration
match.

Other than those things (plus changes to things like the annotate command
which are not used in the match anyway) this is exactly the same search engine
and evaluation as before...

: PS Is Ed allowed to enhance his program during the competition ?

doesn't bother me, although I don't think he has, because he is using Rebel
8. The only change we made to crafty that effected the game was that some
where around move 12 or so, Ed used the "name rebel" command which disables
part of the asymmetric evaluation stuff, and also turns off a critical piece
of code related to timing. Crafty watches how quickly the opponent moves, and
if he moves much faster than crafty is, then it assumes the human is trying to
"flag" it and speeds up. In a handicap match, this can't be done or it will
soon be playing at the same speed as rebel, which would defeat the purpose of
the test...

All of the remainder of the games will be played like this...

Crafty will behave slightly differently, because crafty usse the time control
information to adjust some things internally. For example, the draw score. With
the old time control broken, the draw score was probably "out to lunch" somewhat,
which would make repetitions vary in value from move to move. This won't happen
now, since the time control is (hopefully) staying correct. Again, this has
not affected the NPS game since it is still in game 1... but it will/did affect
the calibration match...


: If yes I assume that the version playing is not the regular (commercial)

Ed Schroder

unread,
Mar 22, 1997, 3:00:00 AM3/22/97
to

From: Patrick Jagstaidt <jags...@iprolink.ch>

: Robert Hyatt wrote:
: [stuff skipped]
: > I have fixed a couple of bugs that have caused a few problems in
: game 1, and will have an executable ready for when game 2 begins.
: [more stuff skipped]

: If you allow for bug fixes between games the whole match becomes a
: moving target. It was already difficult to discriminate what was due to

: 1. Time odd
: 2. Difference in evaluation function efficacy

: but now in between bug fixes are furthermore blurring the issue...

: PS Is Ed allowed to enhance his program during the competition ?

: If yes I assume that the version playing is not the regular (commercial)


: Rebel 8 which is frozen but a sequel, maybe the "enhanced" Rebel 8 under
: beta test on the server...

Bob's bugfix is only meant to solve some AUTO232 problems Crafty is
facing, also Bob implemented a display function so that I can watch
Crafty's time control better and report to Bob in case of problems.

Bob hasn't changed the chess engine.

Also the normal commercial version of Rebel8 is playing. In this way
everbody who owns Rebel8 can check the game. Since Crafty is free
everbody can check the Crafty moves.

Today I already made a second update of the game on my home page. Crafty
has only 6 hours left to the time control and therefore has to move
fast. The game seems to be moving in an ending with material advantage
for Rebel8.

Contents can be found at: http://www.xs4all.nl/~rebchess/match.htm

- Ed Schroder -


: Best regards,
: P. Jagstaidt

Len

unread,
Mar 22, 1997, 3:00:00 AM3/22/97
to

The position after 37. Bc5 is really interesting. After Nxc5 Rxd6,
Crafty cannot recapture the rook because of the mate threat (Qa8). Very
nice. Did Crafty see that it would loose the exchange before it entered
into this line? Did Rebel8 see it?

Anyway, after the forced queen trade, I don't think Crafty will have
enough material left to draw. Crafty can create some threats by pushing
the e and a pawns, but I don't think it’s enough.

Len

Rolf Tueschen

unread,
Mar 22, 1997, 3:00:00 AM3/22/97
to

Does Bob Hyatt give up game one??
~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~~~

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

>Ed Schroder (rebc...@xs4all.nl) wrote:

>: One thing game-1 already proves...
>: In this game Crafty thinks average 4 plies deeper than Rebel due to the
>: 100 x time advantage. Still Crafty is in deep trouble. Bob said Rebel's
>: evaluation function is better than Crafty's evaluation function. I agree
>: with Bob for game-1. Game2-10 will tell us more.

>: Coming to my point: Doesn't game-1 already prove that an evaluation


>: function is (can be) worth many many plies?

>Yes... and don't forget that I've never said otherwise. If you look at my


>original statement "two equivalent programs... a 3-4-5 ply search advantage
>by one is *very* significant."

>The thing I see, which was a minor irritant in 11.17, is that due to a change
>I made, the evaluation for passed pawns went somewhat higher than I planned.

May I dare to remember that we did bet for a match between two well defined
prgs?? For me it sounds as if BH now tried to tune his machine after each
ply/fault to match the dangerous Rebel. Not known before.
Note that I don't object the ameliorating process at all. BH should do that,
but he shouldn't enter after each game done a new version into the ring.
Opinions about that?

> If
>you look at each of the moves Crafty has made so far, they all revolve around the
>passed e pawn. At nearly every move, as it searches deeper than Rebel by a big
>margin, it's found a way to hold that pawn, even if it means giving up something
>somewhere else. And it continued to do that until it was too late to save it,

Bob, we fell with you, understood. But please try to be a fair sportsman. And
look ahead of this single problem of your passed pawn. Next time it will be
another of the many possible attractive goals ...

From a scientifical point you could forfait the match at this moment. We've
already seen that you must tune a lot to meat Rebel's strong attack. I think I
was one of the first who expected this before. Because factor 100 doesn't mean
that much if the underlying strength is so weak. Sorry.
On the other hand in the sportive sense of the event your words sound like a
crying game in the edge of Joe Frazier before another rematch loss against
sting- like- a- bee-Ali. *I am the greatest.*

>I have fixed a couple of bugs that have caused a few problems in game 1,

ok.

>and will
>have an executable ready for when game 2 begins.

No, that doesn't seem correct, Bob. That would be doping.

>The main one is that if the
>game has to be re-started, Crafty won't spend 2x the normal time on the first
>move after the restart, which is a big waste here since the reason for doing this
>is to offset the fact that the trans/ref table has no useful info in it. After
>a huge search it *still* has no useful info in it. :) The next "problem" I
>found is that when game 2 starts, with Crafty using a book

You didn't read my posting that you HAD a book in game 1. Not my fault if you
left it after move 3. :)

>(I hope it seems
>to work this time) it can easily reach move 30 in book. That's going to make
>the last 10 moves in the time control take *forever*.

You can't get enough. A time factor 100 is still too less?

--

>: I am not speaking about Rebel but generally...

>: - Ed Schroder -

Len

unread,
Mar 23, 1997, 3:00:00 AM3/23/97
to

Ed Schroder wrote:

[snip}

> Coming to my point: Doesn't game-1 already prove that an evaluation
> function is (can be) worth many many plies?
>

> I am not speaking about Rebel but generally...
>
> - Ed Schroder -

I have always argued (without much success) that the top programs play
better positionally than most folks here are willing to admit, and I
think this game proves my point. It shows that combinations are based on
positional advantage as much as search depth. A top program must have
the ability to bring about a position that leads to a combination.
Otherwise, it could not beat so many strong humans.

Len

Len

unread,
Mar 23, 1997, 3:00:00 AM3/23/97
to

Robert Hyatt wrote:

[snip]

> In this game, Crafty happens to value passed pawns quite highly, a
> result of tuning to play against humans.

One of the key points in the game is when Rebel8 willingly allowed Crafty
to obtain a passed e pawn by playing 20. dxc3 instead of 20. Bxc3. Why
did Rebel play this move when it seems to violate several positional
rules? It creates a double c pawn, gives up a center pawn, and allows
Crafty to have a passed pawn.

I think the answer is that Rebel really understands the requirements of
the position. In a Bishop vs. Knight game, the side with the Bishop must
restrict the Knight’s mobility by taking away its support points. This
one rule is much more important than the others I previously mentioned.

After 20. dxc3, Rebel controls the important b4 and d4 squares with its
new c pawn. In fact, Rebel controls every square on the 4th rank with
its pawns with the exception of e4. As the game progresses, Crafty never
finds a good post for its Knight, while Rebel’s Bishops turn into
monsters.

Len

Komputer Korner

unread,
Mar 24, 1997, 3:00:00 AM3/24/97
to

Len wrote:
>
> Robert Hyatt wrote:
>
> [snip]

>
>
>
> After 20. dxc3, Rebel controls the important b4 and d4 squares with its
> new c pawn. In fact, Rebel controls every square on the 4th rank with
> its pawns with the exception of e4. As the game progresses, Crafty never
> finds a good post for its Knight, while Rebel’s Bishops turn into
> monsters.
>
> Len

What this match is proving is that if one program is considerably
better than another, then a 3 or 4 ply handicap isn't enough. However
a 6 or 7 ply advantage which is what Deep Blue will have over the
micros would be significant. However Ed has said he needs 3 min.
per move therefore a real test would be a 1000 time handicap or
3000 minutes per move test. That unfortunately would take too long.
Therefore the debate will rage on until Deep Blue decides to play
the micros in a slow time control game.

--
Komputer Korner

The inkompetent komputer.

brucemo

unread,
Mar 24, 1997, 3:00:00 AM3/24/97
to

Komputer Korner wrote:

> What this match is proving is that if one program is considerably
> better than another, then a 3 or 4 ply handicap isn't enough. However
> a 6 or 7 ply advantage which is what Deep Blue will have over the
> micros would be significant. However Ed has said he needs 3 min.
> per move therefore a real test would be a 1000 time handicap or
> 3000 minutes per move test. That unfortunately would take too long.
> Therefore the debate will rage on until Deep Blue decides to play
> the micros in a slow time control game.

With all due respect, it is not proving anything.

If a one-game match is considered proof of anything, then Spassky would
have beaten Fischer in 1972, and Deep Blue would have beaten Kasparov in
their last match.

Crafty is not a cream puff. If Crafty had a graphical interface and cost
fifty bucks, many of you would have bought it and have been satisfied.

Let us continue.

bruce

Robert Hyatt

unread,
Mar 25, 1997, 3:00:00 AM3/25/97
to

Komputer Korner (kor...@netcom.ca) wrote:

: Len wrote:
: >
: > Robert Hyatt wrote:
: >
: > [snip]
: >
: >
: >
: > After 20. dxc3, Rebel controls the important b4 and d4 squares with its
: > new c pawn. In fact, Rebel controls every square on the 4th rank with
: > its pawns with the exception of e4. As the game progresses, Crafty never
: > finds a good post for its Knight, while Rebel’s Bishops turn into
: > monsters.
: >
: > Len

: What this match is proving is that if one program is considerably


: better than another, then a 3 or 4 ply handicap isn't enough. However
: a 6 or 7 ply advantage which is what Deep Blue will have over the
: micros would be significant. However Ed has said he needs 3 min.
: per move therefore a real test would be a 1000 time handicap or
: 3000 minutes per move test. That unfortunately would take too long.
: Therefore the debate will rage on until Deep Blue decides to play
: the micros in a slow time control game.

:

Way too early. if the match reaches 5-0 you might be right. It's not even
1-0 yet... And the result of the match is not the only information we get
here. Will be interesting to see what Crafty sees that Rebel doesn't, and
vice versa...

Robert Hyatt

unread,
Mar 25, 1997, 3:00:00 AM3/25/97
to

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:
: Does Bob Hyatt give up game one??

Bob Hyatt's not playing game one... :)

: ~~~~~~~~~~~~~~~~~~~~~~~~~~
: ~~~~~~~~~~~~~~~~~~~~~~~~~~

: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

: >Ed Schroder (rebc...@xs4all.nl) wrote:

: >: One thing game-1 already proves...
: >: In this game Crafty thinks average 4 plies deeper than Rebel due to the
: >: 100 x time advantage. Still Crafty is in deep trouble. Bob said Rebel's
: >: evaluation function is better than Crafty's evaluation function. I agree
: >: with Bob for game-1. Game2-10 will tell us more.

: >: Coming to my point: Doesn't game-1 already prove that an evaluation


: >: function is (can be) worth many many plies?

: >Yes... and don't forget that I've never said otherwise. If you look at my


: >original statement "two equivalent programs... a 3-4-5 ply search advantage
: >by one is *very* significant."

: >The thing I see, which was a minor irritant in 11.17, is that due to a change
: >I made, the evaluation for passed pawns went somewhat higher than I planned.

: May I dare to remember that we did bet for a match between two well defined
: prgs?? For me it sounds as if BH now tried to tune his machine after each
: ply/fault to match the dangerous Rebel. Not known before.
: Note that I don't object the ameliorating process at all. BH should do that,
: but he shouldn't enter after each game done a new version into the ring.
: Opinions about that?

May I dare suggest that you simply buzz the hell off? If you had followed
the discussion here, you'd certainly know that there were many changes made
to Crafty during this match. And you'd also note that *none* were to the
chess program (engine) itself. No evaluation changes. No search changes.
They were related to two, and only two issues. (1) auto232 has some strange
timing problems that I've been working on with Ed. So far, I don't know if
this is fixed or not. (2) the AN command used by auto232 to start a new game
did not work correctly in Crafty, because Crafty never sees this command in
normal playing on ICC. Ed found the problem, which simply screwed up the
time control for the 2nd-nth game, because things were not reset properly.

Now, with that explanation in plain view (again), I'd be nice if you'd either
first read about what's going on, or else ask someone that knows what's going
on, rather than putting your foot so far into your mouth that you must be
nearly choking on it. The versions Ed has been testing have been identical.
There's been no confusion. This has been stated more than once. Either sit
back and watch, or go somewhere else and do something useful. But stay *out*
of what's going on, since Ed and I are quite old enough and smart enough to
make this work like we want. If you don't like the way we are running it,
go do your own experiment. This one's ours... We don't need someone that
has no idea of what is going on, speculating about things that he knows nothing
about.

The quote above, BTW, is for 11.17, which *is* the version that has been
playing in the match. The passed pawn changes were made *before* anything
was ever sent to Ed. I'd love to scale 'em back down, but I felt that the
program had to be "constant" for the entire match. Excepting for the auto232
stuff which is making the match difficult to run, since every time a game
ends, everything has to be stopped and restarted at present to restore the
time control. Now, with all that said, please go away...

: > If


: >you look at each of the moves Crafty has made so far, they all revolve around the
: >passed e pawn. At nearly every move, as it searches deeper than Rebel by a big
: >margin, it's found a way to hold that pawn, even if it means giving up something
: >somewhere else. And it continued to do that until it was too late to save it,

: Bob, we fell with you, understood. But please try to be a fair sportsman. And
: look ahead of this single problem of your passed pawn. Next time it will be
: another of the many possible attractive goals ...

The passed pawn wasn't the only problem. The point of the above was, and still
is, that at the depth Crafty's been searching, it can pretty much do anything it
wants to. Whether what it wants to do is the right thing to do or not is another
matter, but that's evaluation. The search is doing just exactly what it is
supposed to do here... steer the game toward positions that the eval says are
better. It's done that. The eval was mistaken about the passed pawn. That
happens...

: From a scientifical point you could forfait the match at this moment. We've


: already seen that you must tune a lot to meat Rebel's strong attack. I think I
: was one of the first who expected this before. Because factor 100 doesn't mean
: that much if the underlying strength is so weak. Sorry.

I doubt you'd recognize a weak program if it snuck up and crapped in your
chair. Didn't I recall some odd thing about Stobor vs Crafty from you a while
back, where you sent me two games that crafty won, with the note that "Stobor
kills Crafty"???

I'll post some interesting analysis later, but you might be surprised at
some of what went on in this match. For example. in at least two places,
Crafty played a weaker move, simply because it saw, at depth=15, something
that it thought Rebel would play and it didn't like it. In these two cases
I know of, I've tested rebel on the position and it wouldn't have played the
move Crafty feared. So Crafty simply avoided something that would not have
happened, by playing something that was worse. It was one of the common
scenarios I've seen when two programs play and one outsearches the other by
a significant amount.

: On the other hand in the sportive sense of the event your words sound like a


: crying game in the edge of Joe Frazier before another rematch loss against
: sting- like- a- bee-Ali. *I am the greatest.*

: >I have fixed a couple of bugs that have caused a few problems in game 1,

: ok.

: >and will
: >have an executable ready for when game 2 begins.

: No, that doesn't seem correct, Bob. That would be doping.

Nope. See above. you snipped too much. The bugs were not related to playing
chess, but were *only* related to auto232..


: >The main one is that if the


: >game has to be re-started, Crafty won't spend 2x the normal time on the first
: >move after the restart, which is a big waste here since the reason for doing this
: >is to offset the fact that the trans/ref table has no useful info in it. After
: >a huge search it *still* has no useful info in it. :) The next "problem" I
: >found is that when game 2 starts, with Crafty using a book

: You didn't read my posting that you HAD a book in game 1. Not my fault if you
: left it after move 3. :)

Didn't have a book. The book had permissions set to no write access, so that
Crafty could not open it. It used the "books.bin" file, which has a few dozen
moves in it to kick crafty down the lines I think it plays best, but nothing
more. Is that hard to grasp? If so, how about 1+1=2? maybe you can cope
with that? Crafty has several thousand games that follow the b3 Sicilian
Ed played. Even Moritz was surprised as it was in his book that he sent
Ed. We all forgot about copying a CDROM file which leaves it read-only most
of the time. And crafty has to be able to write into it for the book learning.
When the open failed, it ignored the big book *completely.*

: >(I hope it seems


: >to work this time) it can easily reach move 30 in book. That's going to make
: >the last 10 moves in the time control take *forever*.

: You can't get enough. A time factor 100 is still too less?

I really don't know why I bother responding to you. I'd have more luck
talking to a street lamp. This is the way *all* chess programs I know of
work... if they play several moves out of a book, they get to use that time
when they want? Is that incorrect? If Crafty and Rebel go 30 moves in book,
won't they each have about 3x the normal time to use?


: --

: >: I am not speaking about Rebel but generally...

: >: - Ed Schroder -

: >: Rebel's chances to win this game are looking good but I think

Jmcc28

unread,
Mar 25, 1997, 3:00:00 AM3/25/97
to

In article someone writes:

>: What this match is proving is that if one program is considerably
>: better than another, then a 3 or 4 ply handicap isn't enough. However
>: a 6 or 7 ply advantage which is what Deep Blue will have over the
>: micros would be significant. However Ed has said he needs 3 min.
>: per move therefore a real test would be a 1000 time handicap or
>: 3000 minutes per move test. That unfortunately would take too long.
>: Therefore the debate will rage on until Deep Blue decides to play
>: the micros in a slow time control game.

I just wanted to make a few observations here. Ok just to get this out of
the way I'm no computer programer and I'm not evern all that great at
chess but I do love chess and this is my favorite newsgroup.

From reading this newsgroup one of many big issues was how huch better
Deep Blue was than the other micros. Bob Hyatt took the position that it
was about 200 points higher and Ed thought it was only about 50 points
higher. Mr Hyatt thought Given the massive speed increase in hardware the
only way to explain how deep blue could only be a mere 50 points higher
than the the other micros would be to say that the programer was
incompetent.

Presumably the other side would take the position that IBM's programer
need not be incompetent but nonethless probably doesn't have as good of
evals as the *strongest* micros. they then would argue that this lag in
evals is more important than the increased speed than Bob Hyatt is willing
to concede (thus only a 50 point difference).

Because of the sparse evidence it seems to me both sides have a reasonable
points of view despie the clear disagreement.

Thus the test: take a program(crafty) by someone who is conceded to be at
least competent and play it against the micro that has been topping the
SSDF list (rebel8). And Give Crafty more time to compensate for the
slower hardware.

IMHO this test is helpful to determine who's understanding needs to be
revised more. Nonetheless, the ultimate question that this test was
supposed to provide evidence of is: How much better is Deep Blue than the
best micro? IMHO the following factors shoudl be considered. One factor
suggests that Deep blue woudl be stronger the other suggests that deep
blue woudl be weaker;

1) DB might be stronger than this test would indicate because it will
apparently be even faster than what these time odds represent .(see the
above quote)

but:

2) I think DB may not be as strong as this test could indicate because I
think chances are Craftys evals are probably better.
Let me explain why I think this. Crafty Is considered the *strongest*
free ware available and when you consider all of the software available it
is member of an elite bunch. Now I woudl assume IBM's programer is no
slouch but How sure are we that every piece of software he writes will be
in this elite catagory?

Furthermore (and these are more questions than comments) is it different
writing a program for deep blue than it is for every other micro in the
world? if it is, woudl it be fair to say he can't benifit from common
experiences in the programming world as much as every other software
maker? *If* it is materially different then it seems you have hundreds of
people trying to design software for one system versus one person (or
probably more precisely one team) trying to design software for another
system. Whose basket ball team are you going to bet wins the team that
has the best players from a 2,000 kid highschool or the team that only has
25 kids?

Whether Crafty put too much weight on the past pawn or not is not really
relevant. My point is that DB will undoubtedly make similar sorts of
misevaluations. And *if* what I was saying above has any truth to it DB
will probably have more than Crafty which has been played and tested by
people and computers all over the world. In short, Just because crafty
misevaluated the passed pawn doesn't mean that its programmer is
incompetent. Nobody can say exactly what it is that Deep blue will
misevaluate but given statistics the misevaluations should be there and
probably in greater numbers.
As far as crafty being programmed to beat humans I would imagine that the
DB team is also programming to beat humans and this too may put it at a
disadvantage against micros. But its still a disadvantage.
Finnally Just because I would buy a faster cpu given the chance that
doesn't mean that I think that speed of the cpu is all thatimportant *at
tournamnt time controls*. The reason I would want a fast cpu is not
because I think I will have a much better opponent for tounament time
control games. Remeber that the argument goes that once you get highter
and higher in ply there is a diminished return. Thus I would get the
faster cpubecause i don't like to wait 4 hours to get an analysis of a
game I just played. I want to get an analyis preferably in minutes.
given that the speed of the cpu has the biggest impact at these shorter
time periods I woudl want the fastest cpu so that I wouldn't need to wait
as long for good ananlysis.

Anyways the match is far from over and it is great to watch. My thanks to
both Bob Hyatt and Ed Schroeder!

Joe McCarron

Robert Hyatt

unread,
Mar 25, 1997, 3:00:00 AM3/25/97
to

Jmcc28 (jmc...@aol.com) wrote:
: In article someone writes:

: but:

Pretty sure, unfortunately. I've known most of the "group" for many
years now. Murray the longest, but Hsu and company as well. There's nothing
I know of to suggest that what they are doing in DB is inferior to what I
do in Crafty. Some things are grossly different, in that they don't do any
of the risky things I do (null move, razoring, etc.) because they can *still*
reach a superior search depth without those tricks.

As far as the eval goes, DB probably does as much or more than I am doing,
from conversations with Hsu... but I'll leave the details for him to explain
when he's ready...


: Furthermore (and these are more questions than comments) is it different


: writing a program for deep blue than it is for every other micro in the
: world? if it is, woudl it be fair to say he can't benifit from common
: experiences in the programming world as much as every other software
: maker? *If* it is materially different then it seems you have hundreds of
: people trying to design software for one system versus one person (or
: probably more precisely one team) trying to design software for another
: system. Whose basket ball team are you going to bet wins the team that
: has the best players from a 2,000 kid highschool or the team that only has
: 25 kids?

It's not that different, but the parallel search is different. The main
difference is that Hsu and company have published details about what they
are doing. As have I. As has Jonathan Schaeffer. etc. Not to forget
the "mother of all chess articles" by David Slate. Everyone has used the
ideas from all of the above, but there is one particular group that is
noticably silent when it's time for the information to flow the other way.

*noticably* silent...


: Whether Crafty put too much weight on the past pawn or not is not really

mclane

unread,
Mar 26, 1997, 3:00:00 AM3/26/97
to

brucemo <bru...@nwlink.com> wrote:

>Komputer Korner wrote:

>> What this match is proving is that if one program is considerably
>> better than another, then a 3 or 4 ply handicap isn't enough. However
>> a 6 or 7 ply advantage which is what Deep Blue will have over the
>> micros would be significant. However Ed has said he needs 3 min.
>> per move therefore a real test would be a 1000 time handicap or
>> 3000 minutes per move test. That unfortunately would take too long.
>> Therefore the debate will rage on until Deep Blue decides to play
>> the micros in a slow time control game.

>With all due respect, it is not proving anything.

>If a one-game match is considered proof of anything, then Spassky would
>have beaten Fischer in 1972, and Deep Blue would have beaten Kasparov in
>their last match.

>Crafty is not a cream puff. If Crafty had a graphical interface and cost
>fifty bucks, many of you would have bought it and have been satisfied.

>Let us continue.

>bruce
Why do you always get offended by someones post, you don't have to
defend Crafty. Of course we like it, and we would even buy it having
no I/O-interface, maybe not for 50$ but for shareware-prices. But
please try to calm down more. In europe having different opinion and
TELLING this opinion to another person does not include that you
a) have something against this person
b) you hate him
c) you want to destroy or kill him
d) you don't like him.

Different opinon means only different opinon. I often have the feeling
that you always take anything that is said here to personal.

Maybe you will now call this an attack. It isn't. I thought you are a
nice guy, since I met you the first time in Paderborn. Although you
maybe always thought I have something against YOU. Whatever.

Back to the topic: Of course a few samples are not that much an
evidence for anything. BUT WE STUDY NOT ONLY THE GAMES and see the
end-result, we also study the evaluations and the main-lines. And this
IS OF COURSE the main information that helps us to qualify these
quantity of events (each game one event with many subevents [the moves
in each position]) .

If you ONLY look for the results of an experiment with 5 events, than
it would be say nothing. But if you also look for each subenevent
(each position) that you have in 5 games maybe 250 events. That is a
number of events that says much, and that we have much time to study
the evaluations of the programs is also one main goal that helps to
us.

I mean, that is how I test chess-programs. I look the games, study the
main-lines. Try out and set-up positions and after I have an idea I
try to prove my suspicions with autoplayer-results or special-created
(artificial) positions.


brucemo

unread,
Mar 26, 1997, 3:00:00 AM3/26/97
to

Robert Hyatt wrote:

> I doubt you'd recognize a weak program if it snuck up and crapped in your
> chair. Didn't I recall some odd thing about Stobor vs Crafty from you a while
> back, where you sent me two games that crafty won, with the note that "Stobor
> kills Crafty"???

Oh god no. That was Thorsten, not Tueschen. Let's not go down that road again.

bruce

Robert Hyatt

unread,
Mar 26, 1997, 3:00:00 AM3/26/97
to

brucemo (bru...@nwlink.com) wrote:
: Robert Hyatt wrote:

: > I doubt you'd recognize a weak program if it snuck up and crapped in your


: > chair. Didn't I recall some odd thing about Stobor vs Crafty from you a while
: > back, where you sent me two games that crafty won, with the note that "Stobor
: > kills Crafty"???

: Oh god no. That was Thorsten, not Tueschen. Let's not go down that road again.

: bruce

My mistake. and I stand corrected. Difficult to tell them apart at times,
at other times it is easy. My appologies...

Enrique Irazoqui

unread,
Mar 26, 1997, 3:00:00 AM3/26/97
to

brucemo <bru...@nwlink.com> escribió en artículo
<333731...@nwlink.com>...

> Komputer Korner wrote:
>
> > What this match is proving is that if one program is considerably
> > better than another, then a 3 or 4 ply handicap isn't enough. However
> > a 6 or 7 ply advantage which is what Deep Blue will have over the
> > micros would be significant. However Ed has said he needs 3 min.
> > per move therefore a real test would be a 1000 time handicap or
> > 3000 minutes per move test. That unfortunately would take too long.
> > Therefore the debate will rage on until Deep Blue decides to play
> > the micros in a slow time control game.
>
> With all due respect, it is not proving anything.
>
> If a one-game match is considered proof of anything, then Spassky would
> have beaten Fischer in 1972, and Deep Blue would have beaten Kasparov in
> their last match.

They didn't play at 100:1, you shouldn't forget this. It's as if Genius 5,
as far as I know the best at blitz, would beat Rebel 8 at 100:1. Say, G5 at
2 seconds and R8 at 200 seconds/move. I'm quite sure that in a 10 games
match Genius 5 couldn't win one single game. Rebel 8 did win at 100:1. It
must mean something. Better evaluations to begin with. And in this
particular game the better evals were more important than a 4 ply deeper
search. Only one game, I know, but a very significant game.

> Crafty is not a cream puff. If Crafty had a graphical interface and cost

> fifty bucks, many of you would have bought it and have been satisfied.

With all due respect, this is not proving anything.

Enrique

> Let us continue.
>
> bruce
>

Douglas Crowe

unread,
Mar 26, 1997, 3:00:00 AM3/26/97
to

On 22 Mar 1997 07:26:28 GMT, TUESCHEN.MEDIZ...@t-online.de
(Rolf Tueschen) wrote:

>Does Bob Hyatt give up game one??

>~~~~~~~~~~~~~~~~~~~~~~~~~~
>~~~~~~~~~~~~~~~~~~~~~~~~~~
>
>hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>
>>Ed Schroder (rebc...@xs4all.nl) wrote:
>
>>: One thing game-1 already proves...
>>: In this game Crafty thinks average 4 plies deeper than Rebel due to the
>>: 100 x time advantage. Still Crafty is in deep trouble. Bob said Rebel's
>>: evaluation function is better than Crafty's evaluation function. I agree
>>: with Bob for game-1. Game2-10 will tell us more.
>
>>: Coming to my point: Doesn't game-1 already prove that an evaluation
>>: function is (can be) worth many many plies?
>
>>Yes... and don't forget that I've never said otherwise. If you look at my
>>original statement "two equivalent programs... a 3-4-5 ply search advantage
>>by one is *very* significant."
>
>>The thing I see, which was a minor irritant in 11.17, is that due to a change
>>I made, the evaluation for passed pawns went somewhat higher than I planned.
>
>May I dare to remember that we did bet for a match between two well defined
>prgs?? For me it sounds as if BH now tried to tune his machine after each
>ply/fault to match the dangerous Rebel. Not known before.
>Note that I don't object the ameliorating process at all. BH should do that,
>but he shouldn't enter after each game done a new version into the ring.
>Opinions about that?
>

>> If
>>you look at each of the moves Crafty has made so far, they all revolve around the
>>passed e pawn. At nearly every move, as it searches deeper than Rebel by a big
>>margin, it's found a way to hold that pawn, even if it means giving up something
>>somewhere else. And it continued to do that until it was too late to save it,
>
>Bob, we fell with you, understood. But please try to be a fair sportsman. And
>look ahead of this single problem of your passed pawn. Next time it will be
>another of the many possible attractive goals ...
>

>From a scientifical point you could forfait the match at this moment. We've
>already seen that you must tune a lot to meat Rebel's strong attack. I think I
>was one of the first who expected this before. Because factor 100 doesn't mean
>that much if the underlying strength is so weak. Sorry.

>On the other hand in the sportive sense of the event your words sound like a
>crying game in the edge of Joe Frazier before another rematch loss against
>sting- like- a- bee-Ali. *I am the greatest.*
>
>>I have fixed a couple of bugs that have caused a few problems in game 1,
>
>ok.
>
>>and will
>>have an executable ready for when game 2 begins.
>
>No, that doesn't seem correct, Bob. That would be doping.
>

>>The main one is that if the
>>game has to be re-started, Crafty won't spend 2x the normal time on the first
>>move after the restart, which is a big waste here since the reason for doing this
>>is to offset the fact that the trans/ref table has no useful info in it. After
>>a huge search it *still* has no useful info in it. :) The next "problem" I
>>found is that when game 2 starts, with Crafty using a book
>
>You didn't read my posting that you HAD a book in game 1. Not my fault if you
>left it after move 3. :)
>

>>(I hope it seems
>>to work this time) it can easily reach move 30 in book. That's going to make
>>the last 10 moves in the time control take *forever*.
>
>You can't get enough. A time factor 100 is still too less?
>

>--
>
>>: I am not speaking about Rebel but generally...
>
>>: - Ed Schroder -
>
>>: Rebel's chances to win this game are looking good but I think
>>: Crafty can hold the king attack and perhaps force an ending with one
>>: (or two?) pawns down. Game-1 is certainly not over yet.
>
>
>
>

Excuse me for interupting,but is there a place where this
game between Crafty and Rebel can be followed,or is it
just updated here in the group?Thanks in advance.

Douglas Crowe

jmc...@aol.com

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

In article someone writes:

>: What this match is proving is that if one program is considerably


>: better than another, then a 3 or 4 ply handicap isn't enough. However
>: a 6 or 7 ply advantage which is what Deep Blue will have over the
>: micros would be significant. However Ed has said he needs 3 min.
>: per move therefore a real test would be a 1000 time handicap or
>: 3000 minutes per move test. That unfortunately would take too long.
>: Therefore the debate will rage on until Deep Blue decides to play
>: the micros in a slow time control game.

I just wanted to make a few observations here. Ok just to get this out of the way I'm no computer programer and I'm not evern all that great at chess but I do love chess and this is my favorite newsgroup.

From reading this newsgroup one of many big issues was how huch better Deep Blue was than the other micros. Bob Hyatt took the position that it was about 200 points higher and Ed thought it was only about 50 points higher. Mr Hyatt thought Given the massive speed increase in hardware the only way to explain how deep blue could only be a mere 50 points higher than the the other micros would be to say that the programer was incompetent.

Presumably the other side would take the position that IBM's programer need not be incompetent but nonethless probably doesn't have as good of evals as the *strongest* micros. they then would argue that this lag in evals is more important than the increased speed than Bob Hyatt is willing to concede (thus only a 50 point difference).

Because of the sparse evidence it seems to me both sides have a reasonable points of view despie the clear disagreement.

Thus the test: take a program(crafty) by someone who is conceded to be at least competent and play it against the micro that has been topping the SSDF list (rebel8). And Give Crafty more time to compensate for the slower hardware.

IMHO this test is helpful to determine who's understanding needs to be revised more. Nonetheless, the ultimate question that this test was supposed to provide evidence of is: How much better is Deep Blue than the best micro? IMHO the following factors shoudl be considered. One factor suggests that Deep blue woudl be stronger the other suggests that deep blue woudl be weaker;

1) DB might be stronger than this test would indicate because it will apparently be even faster than what these time odds represent .(see the above quote)

but:

2) I think DB may not be as strong as this test could indicate because I think chances are Craftys evals are probably better.
Let me explain why I think this. Crafty Is considered the *strongest* free ware available and when you consider all of the software available it is member of an elite bunch. Now I woudl assume IBM's programer is no slouch but How sure are we that every piece of software he writes will be in this elite catagory?

Furthermore (and these are more questions than comments) is it different writing a program for deep blue than it is for every other micro in the world? if it is, woudl it be fair to say he can't benifit from common experiences in the programming world as much as every other software maker? *If* it is materially different then it seems you have hundreds of people trying to design software for one system versus one person (or probably more precisely one team) trying to design software for another system. Whose basket ball team are you going to bet wins the team that has the best players from a 2,000 kid highschool or the team that only has 25 kids?

Whether Crafty put too much weight on the past pawn or not is not really relevant. My point is that DB will undoubtedly make similar sorts of misevaluations. And *if* what I was saying above has any truth to it DB will probably have more than Crafty which has been played and tested by people and computers all over the world. In short, Just because crafty misevaluated the passed pawn doesn't mean that its programmer is incompetent. Nobody can say exactly what it is that Deep blue will misevaluate but given statistics the misevaluations should be there and probably in greater numbers.

Komputer Korner

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

Douglas Crowe wrote

> >
> >
> >
> >
> Excuse me for interupting,but is there a place where this
> game between Crafty and Rebel can be followed,or is it
> just updated here in the group?Thanks in advance.
>
> Douglas Crowe

http://www.xs4all.nl/~rebchess/

Christopher R. Dorr

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to


Rolf Tueschen <TUESCHEN.MEDIZ...@t-online.de> wrote in article
<5h01j4$1...@news00.btx.dtag.de>...


> Does Bob Hyatt give up game one??
> ~~~~~~~~~~~~~~~~~~~~~~~~~~
> ~~~~~~~~~~~~~~~~~~~~~~~~~~
>

> From a scientifical point you could forfait the match at this moment.
We've
> already seen that you must tune a lot to meat Rebel's strong attack. I
think I
> was one of the first who expected this before. Because factor 100 doesn't
mean
> that much if the underlying strength is so weak. Sorry.
> On the other hand in the sportive sense of the event your words sound
like a
> crying game in the edge of Joe Frazier before another rematch loss
against
> sting- like- a- bee-Ali. *I am the greatest.*
>

Forfeit the match after one game? That's absurd. One game proves exactly
nothing. It may be suggestive of something, but if you feel that this
proves something one way or the other, I can recommend a number of good
*introductory* books on statistics.

Crafty is anything but weak. There is tremendous evidence to support this.
FICS, ICC, the equal time match vs. Rebel, etc...

Also, I'd suggest you check your dictionary, and compare the meaning of
'excuse' versus 'explanation'.

I've never seen Bob making excuses for anything. He's here (as many, but
apparently not all, of us are) to explore ideas in computer chess. He WROTE
Crafty, so it logically follows that he might have some good ideas (perhaps
better than yours?) about what is going on. If this match is going to have
any meaning, it has to include some insight given by those who have a deep
understanding of what is happening, as well as a raw score result.

He has said numerous times that he is not tuning the engine AT ALL. He is
attempting to rectify problems with the autoplay interface.

Why are these things so hard to understand?

Chris


mclane

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

brucemo <bru...@nwlink.com> wrote:

>Robert Hyatt wrote:

>> I doubt you'd recognize a weak program if it snuck up and crapped in your
>> chair. Didn't I recall some odd thing about Stobor vs Crafty from you a while
>> back, where you sent me two games that crafty won, with the note that "Stobor
>> kills Crafty"???

Where I have sent 2 games where CRAFTY won, with the note that Stobor
kills crafty ??!

I think this cannot be true either. Tom had a version (I still have
the executable) that is really strong and will/would kill crafty.
There is no doubt about this fact.

I don't understand why now crafty killed stobor. I though it was vice
versa.


>Oh god no. That was Thorsten, not Tueschen. Let's not go down that road again.

>bruce

Do you really believe that I would have posted stobor kills crafty if
I would not be sure about this ?!
I have both programs, I have machines to find it out.
Nothing is easier. You don't need 100 or 200 games to show evidence.
If you have eyes to see and feelings a small sample says more to you
than any autoplayer can find out in 50 or more games.

I don't understand why you have problems to believe that feelings can
replace statistic perfectly. Don't you love your wife although never
any evidence and statistic was made about your love ?!

BTW: The poster was mclane !

Robert Hyatt

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

mclane (mcl...@prima.ruhr.de) wrote:
: brucemo <bru...@nwlink.com> wrote:

: >Robert Hyatt wrote:

: >> I doubt you'd recognize a weak program if it snuck up and crapped in your


: >> chair. Didn't I recall some odd thing about Stobor vs Crafty from you a while
: >> back, where you sent me two games that crafty won, with the note that "Stobor
: >> kills Crafty"???

: Where I have sent 2 games where CRAFTY won, with the note that Stobor
: kills crafty ??!

: I think this cannot be true either. Tom had a version (I still have
: the executable) that is really strong and will/would kill crafty.
: There is no doubt about this fact.

There's lots of doubt, Crafty's played stobor many times on both ICC and
Chess.net... log on to ICC and "search +=crafty -=ratbert" and then
"search -=crafty +=ratbert"

The first command produces 65 wins for crafty... the second produces 10
losses for crafty...

As the saying goes,

Q.E.D.


: I don't understand why now crafty killed stobor. I though it was vice
: versa.

you sent me 3 games. the first, "stobor kills crafty, *but* stobor made
a mistake in the endgame and lost." the second "stobor kills crafty, *but*
stobor didn't realize that it had to stop the a-pawn from advancing and it
lost." the third "stobor kills crafty, but crafty was lucky and found a
repetition draw."

Those were what I was talking about. It sounds like the old elementary
school conversation, where a kid comes home with a black eye, blood all
over his shirt. His dad asked "did you get into a fight?" The kid
responds, eyes down, "yes.". Dad says "how'd you do?" Kid responds,
"I beat the hell out of his fist with my face." :)

: >Oh god no. That was Thorsten, not Tueschen. Let's not go down that road again.

: >bruce

: Do you really believe that I would have posted stobor kills crafty if
: I would not be sure about this ?!
: I have both programs, I have machines to find it out.
: Nothing is easier. You don't need 100 or 200 games to show evidence.
: If you have eyes to see and feelings a small sample says more to you
: than any autoplayer can find out in 50 or more games.

Exactly. I have enough evidence to convince myself.

: I don't understand why you have problems to believe that feelings can


: replace statistic perfectly. Don't you love your wife although never
: any evidence and statistic was made about your love ?!

: BTW: The poster was mclane !

I corrected it and appologized after Bruce pointed out the error. :)

mclane

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

"Christopher R. Dorr" <crd...@cinternet.net> wrote:

>Rolf Tueschen <TUESCHEN.MEDIZ...@t-online.de> wrote in article
><5h01j4$1...@news00.btx.dtag.de>...
>> Does Bob Hyatt give up game one??
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~
>>
>> From a scientifical point you could forfait the match at this moment.
>We've
>> already seen that you must tune a lot to meat Rebel's strong attack. I
>think I
>> was one of the first who expected this before. Because factor 100 doesn't
>mean
>> that much if the underlying strength is so weak. Sorry.
>> On the other hand in the sportive sense of the event your words sound
>like a
>> crying game in the edge of Joe Frazier before another rematch loss
>against
>> sting- like- a- bee-Ali. *I am the greatest.*
>>

>Forfeit the match after one game? That's absurd. One game proves exactly
>nothing. It may be suggestive of something, but if you feel that this
>proves something one way or the other, I can recommend a number of good
>*introductory* books on statistics.

Statistic proves nothing ! A chess program consists of positions. Each
position can be evaluated. The programs give an evaluation, and they
search. Looking on each evaluation and on each main line gives you (in
a normal game of 50 moves) 100 events where you can SEE and this way
proof if a program plays good chess or not.

Of course - if you only look for the result in the end of the game,
you need more events. But why should we do this ?
I mean, we have eyes to see and the programs have INFORMATION-windows
to show us. Why should we close our eyes, wait 50 moves without doing
anything, take out a book about statistics in the end, and raise on
finger in the air and say: I have found out as a statistician that
this single result proves NOTHING.

I mean: what are your guys doing there ?! If chess is only the result
at the end of the game, why the bloody hell do so many people follow
chess-games and study analysis and thousands of books are written
about if just the result at the end of the game is that important.


No - statistic may be nice, but only fools make 1 event out of 100 !
These fools are called : statisticians !

I give back your own words:

Robert Hyatt

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

Christopher R. Dorr (crd...@cinternet.net) wrote:


: Rolf Tueschen <TUESCHEN.MEDIZ...@t-online.de> wrote in article
: <5h01j4$1...@news00.btx.dtag.de>...
: > Does Bob Hyatt give up game one??
: > ~~~~~~~~~~~~~~~~~~~~~~~~~~
: > ~~~~~~~~~~~~~~~~~~~~~~~~~~
: >
: > From a scientifical point you could forfait the match at this moment.
: We've
: > already seen that you must tune a lot to meat Rebel's strong attack. I
: think I
: > was one of the first who expected this before. Because factor 100 doesn't
: mean
: > that much if the underlying strength is so weak. Sorry.
: > On the other hand in the sportive sense of the event your words sound
: like a
: > crying game in the edge of Joe Frazier before another rematch loss
: against
: > sting- like- a- bee-Ali. *I am the greatest.*
: >

: Forfeit the match after one game? That's absurd. One game proves exactly
: nothing. It may be suggestive of something, but if you feel that this
: proves something one way or the other, I can recommend a number of good
: *introductory* books on statistics.

: Crafty is anything but weak. There is tremendous evidence to support this.


: FICS, ICC, the equal time match vs. Rebel, etc...

: Also, I'd suggest you check your dictionary, and compare the meaning of
: 'excuse' versus 'explanation'.

: I've never seen Bob making excuses for anything. He's here (as many, but
: apparently not all, of us are) to explore ideas in computer chess. He WROTE
: Crafty, so it logically follows that he might have some good ideas (perhaps
: better than yours?) about what is going on. If this match is going to have
: any meaning, it has to include some insight given by those who have a deep
: understanding of what is happening, as well as a raw score result.

: He has said numerous times that he is not tuning the engine AT ALL. He is
: attempting to rectify problems with the autoplay interface.

: Why are these things so hard to understand?

: Chris


While I always appreciate "kind words" since they are pretty rare here, I'd like to
suggest that everyone take a few deep breaths here. This isn't the end of the world.
For those that are interested in this issue, we'll find a way to answer the question
one way or another. For those that want to know how crafty will do "heads up" with
Rebel on equal hardware, the SSDF will eventually answer that whether I like the
answer or not.

Here's a brief description of what went wrong in the match:

1. The AN command is used by auto232 to start a new game. On the servers, xboard/
winboard simply kills the old crafty process and starts a new one after each game.
Auto232 tells crafty to reset and start again. I improperly reset the time control
information, which caused two problems, one of which Ed caught and reported to me.

(a) crafty used bogus time control stuff after the first game in the calibration
match. In some games, it took too long. In others it moved quicker than it
should have. None of this was Ed's fault, it was an error in my code that had
been undetected for well over a year because this code is not used the way I
run crafty on servers.

(b) Crafty also used a bogus time control for the opponent, and this did have some
impact on the game. After the first game, it thought the opponent had zero
time left. If you've watched crafty play on ICC, you've noticed that draw
scores range from 0.000 all the way down to -1.000, depending on the stage
of the game (in part) but more importantly, depending on how much time the
opponent has left on the clock. The problem is, of course, that if Rebel
thought it was slightly worse, it could try to repeat a position to make crafty
move elsewhere. Crafty would try to avoid the repetition, because it would appear
that the repetition was losing a pawn (-1.000 for the evaluation). And the result
is that rather than take a draw, Crafty would likely take a loss, because by the time
the eval got to -1, Rebel would have decided it didn't want the draw after all. :)
This was my error. When I looked at the logs, it became obvious what I'd done. I
never thought the match should be re-started, because if we restart every time I find
a bug, it might never end. :) That was my decision, not Ed's.

(c) in the NPS game, crafty played without an opening book. It appears that when the book
was copied in from CDROM, the permissions were wrong, Crafty couldn't write to it (which
book learning requires) and so it ignored it. The position after move 12 or so looked
perfectly solid to me, and from the point where this occurred, I've always maintained that
not having a book should not have been an issue here. 15 plies is enough to find decent
opening moves as the game shows. So the book might have helped, because the position
would clearly have been different since crafty doesn't play g6 in Sicilian defenses as
black. But also, just as clearly, Crafty got itself into whatever it got into, all by
itself. I thought the game was valid, and never said otherwise.

(d) numerous auto232 bugs were preventing the calibration match from running smoothly. There
are some rediculous timing considerations necessary to make this work right. Crafty can't
wait too long to echo the move the opponent made. It can't echo the move too quickly
either. It has to be "just right" which is not easy since I don't have the facility to
test the auto232 stuff here. This was resulting in many games not getting completed
due to auto232 getting ticked and hanging.

I made one change to fix (a) and (b) above as soon as we found it. I made many changes to
try to solve (d). The engine nor eval were changed at all, although to be technical, the eval
was changed by (b) since it quit seeing bad draw scores after the fix. I left the bogus passed
pawn values alone, even though I'd have loved to change them.

I was less interested in how Crafty did positionally against Rebel than I was/am about how
it would do tactically with the extra search depth. That's still the point I'd like to resolve,
as it's an interesting point for discussion...

With that said, every problem above was mine. If Ed "had enough" that's certainly his
choice to make. It was taking time, the email was flowing in a large volume daily as Bruce
mentioned. I felt/feel bad about the problems, but don't know what I could have done to make
this go any smoother. Difficult to write code for an interface that is *very* touchy, particularly
when I don't have the thing to test it...


Torstein Hall

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

mclane <mcl...@prima.ruhr.de> skrev i artikkelen
<E7pKz...@news.prima.ruhr.de>...


> "Christopher R. Dorr" <crd...@cinternet.net> wrote:
>
>
>
> >Rolf Tueschen <TUESCHEN.MEDIZ...@t-online.de> wrote in
article
> ><5h01j4$1...@news00.btx.dtag.de>...
> >> Does Bob Hyatt give up game one??
> >> ~~~~~~~~~~~~~~~~~~~~~~~~~~
> >> ~~~~~~~~~~~~~~~~~~~~~~~~~~
> >>
> >> From a scientifical point you could forfait the match at this moment.
> >We've
> >> already seen that you must tune a lot to meat Rebel's strong attack. I
> >think I
> >> was one of the first who expected this before. Because factor 100
doesn't
> >mean
> >> that much if the underlying strength is so weak. Sorry.
> >> On the other hand in the sportive sense of the event your words sound
> >like a
> >> crying game in the edge of Joe Frazier before another rematch loss
> >against
> >> sting- like- a- bee-Ali. *I am the greatest.*
> >>
>
> >Forfeit the match after one game? That's absurd. One game proves exactly
> >nothing. It may be suggestive of something, but if you feel that this
> >proves something one way or the other, I can recommend a number of good
> >*introductory* books on statistics.
>

> Statistic proves nothing ! A chess program consists of positions. Each
> position can be evaluated. The programs give an evaluation, and they
> search. Looking on each evaluation and on each main line gives you (in
> a normal game of 50 moves) 100 events where you can SEE and this way
> proof if a program plays good chess or not.

Why are you always wrong? :)

Statistics prove a lot, within certain limits.
Since chess consists out a very big number of positions, its impractical to
test them all! You rather play some games, study the eval of the programs,
and the results of the game and make your conclusion about most of the
positions.
Then you say with that one program is stronger than the other, within your
margin of error. If you can calculate how big the probability of you beeing
right or wrong is, then its statistics!


> Of course - if you only look for the result in the end of the game,
> you need more events. But why should we do this ?
> I mean, we have eyes to see and the programs have INFORMATION-windows
> to show us. Why should we close our eyes, wait 50 moves without doing
> anything, take out a book about statistics in the end, and raise on
> finger in the air and say: I have found out as a statistician that
> this single result proves NOTHING.
>
> I mean: what are your guys doing there ?! If chess is only the result
> at the end of the game, why the bloody hell do so many people follow
> chess-games and study analysis and thousands of books are written
> about if just the result at the end of the game is that important.
>
>
> No - statistic may be nice, but only fools make 1 event out of 100 !
> These fools are called : statisticians !
>
> I give back your own words:

> >Why are these things so hard to understand?
>
> >Chris
>
>
>

--
Torstein Hall
tors...@eunet.no
http://login.eunet.no/torshall/

Rolf Tueschen

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

"Christopher R. Dorr" <crd...@cinternet.net> wrote:

>Rolf Tueschen <TUESCHEN.MEDIZ...@t-online.de> wrote in article
><5h01j4$1...@news00.btx.dtag.de>...
>> Does Bob Hyatt give up game one??
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~
>>
>> From a scientifical point you could forfait the match at this moment.

Note, I wrote this long before Ed Schroder wrote about his emotions when he
read Bob's comments about whats all about in game one. See this in the other
thread. Ed also felt his point already been proven.

>>We've
>> already seen that you must tune a lot to meat Rebel's strong attack. I
>think I
>> was one of the first who expected this before. Because factor 100 doesn't
>mean
>> that much if the underlying strength is so weak. Sorry.
>> On the other hand in the sportive sense of the event your words sound
>like a
>> crying game in the edge of Joe Frazier before another rematch loss
>against
>> sting- like- a- bee-Ali. *I am the greatest.*
>>

>Forfeit the match after one game? That's absurd.

I don't think so.

>One game proves exactly
>nothing.

Pretending with *exactly* a final scientifical result you were caught by sort
of wishful thinking, no?

>It may be suggestive of something, but if you feel that this
>proves something one way or the other, I can recommend a number of good
>*introductory* books on statistics.

Ok. None would build statistically relevance on one single game. I always
opposed this and was informed by a certain SSDF expert that single matches of
three or maybe 1 game wouldn't hurt the overall statistical result -- but I
never bought this, because for me it was putting apples and beans together ...

BUT in this case of resulting game between Crafty factor 100 (!) and Rebel game
one already showed a lot concerning chess content. Or did you believe before
the start that Rebel would set a classical knocker to mute krafti which always
*looked deeper for 3 or four plies*???

Crafty failed to understand a lot of things in this game. I'm not in the
position to explain all of them. I'm a too weak chess player. But without
having read master analysis about the game I'll try as follows:

Crafty was lost with a weak f-file. And a weak f7. Then Rebel decided to change
play from pure mating threatening to the point d5. With the B on c4 seeing at
e6--f7. The rook on f1 --- f7. The Q on f3 against d5 and f7. Then crafty
overlooked in spite of his greater depth the b5 *sac*. Right, the move before
everyone *sees* the move Bxb5, but then it's too late. The whole position is
positionally lost. Normally one had expected theat Crafty should foresee the
positions after Rebel's 33.Bd3. Remember Crafty had material equality at that
moment, but was positionally already lost!

Now the thing Bob did was to go way more backward. He even said that after 10
moves Rebel had a better / won? (I can't remember it right) position. And he
simply argued that the missing book was the source of all that ...

Look, that's what I criticised in a former post. If you argue like that you
could wipe off all the brilliancy and magical of every single won game of
Mikhail Tal for instance. You simple argue backwards having the later results
in mind. And you *show* that after this or that Tal never had the chance to sac
this or that.

If you continue for some time on this -- you'll end in writing books about 1.e4
and wins. Or 1.d4 and mate in 45 moves. Or you could say, that you end up with
playing against god if you start playing chess. It's the same with the egg and
the duck. Or what, if our parents wouldn't have had such a ball we all wouldn't
be here ....


For me at least Rebel won a nice game in a style Kasparov couldn't do better
against Deep Blue. So we had in nucleus what will happen in May. I mean, it
became quite clear that Rebel had enough chess wisdom more than Crafty to
equalize/neutralize the exhorbitant time factor 100.

Don't you agree on this?

Ok, and I'll agree that this single game didn't prove anything scientifically.
but 10 games wouldn't be enough either. I'm even not sure that this sort of
experimental setting was best to test the thesis of both sides. For me Ed's
argument was clear in advance. And the game demonstated that to omit the
following *weaknesses* Crafty had to look 20 not 2 or three moves *more* ahead
than Rebel or just nothing more depth related -- sounds illogical but seems to
hit the truth. For Rebel it was enough to play at brute force level 8/9 to
build a nice attack and afterwards a positional winning. But as far as
Kasparov's level is concerned they both would be without any chances, because
for GK all the discussed wouldn't be primarily a question of search depth but
wisdom of patterns. These are out of sight even for Deep Blue.

>Crafty is anything but weak. There is tremendous evidence to support this.
>FICS, ICC, the equal time match vs. Rebel, etc...

None said the opposite. And Bob's *foreseeing* feature is really dangerous on a
rapid game, because it reacts rather strong on the basis of deep search if
you're caught in a foreseen line. Then a computer is always better than a human
player. That's all.

>Also, I'd suggest you check your dictionary, and compare the meaning of
>'excuse' versus 'explanation'.

>I've never seen Bob making excuses for anything.

How comes? I've read it several times. :)

> He's here (as many, but
>apparently not all, of us are) to explore ideas in computer chess. He WROTE
>Crafty, so it logically follows that he might have some good ideas (perhaps
>better than yours?)

We agree 100% on that one for sure. :)

>about what is going on. If this match is going to have
>any meaning, it has to include some insight given by those who have a deep
>understanding of what is happening, as well as a raw score result.

Yes, I agree again. But I still want to differentiate the programming stuff
(here ES is the suitable side) and tthe chess stuff. And that would lead us
back to the Kasparov vs Deep Blue match. I think in repetition that this match
would be the continuation of the picture one could see in game one discussed
above.

>He has said numerous times that he is not tuning the engine AT ALL. He is
>attempting to rectify problems with the autoplay interface.

>Why are these things so hard to understand?

Because it's not true. If I had the time I could quote Bob's writing about
several versions in the next 6 months. And he surely wanted to *tune*. The
book, the passed pawn stuff and so on. You must read a bit more carefully. But
I agree Bob did never propose to enter version 11.19 or 20 for game 2 or 3. But
all the wining meant exactly the same (implications wise ...). Let me even go
further to baite Bob. He dreamed of a tuned version which had learned enough
about Rebel to master his slave thereafter. And I'm even sure that this would
be possible to implantate.
Bob, I'm not doing anything spooky. It's only exact (!) reading of your
writings. (:-))={ I don't know, but I always hit the ceiling these days.
What's this all about? I can really fly. Me thinks.

>Chris

Greetings and thanks

Rolf

P.S. Remember I gave my betting at the beginning of the game with 6 to 7 points
FOR Rebel as the winner. Maybe that was for a great joy all over this planet
..

But I did this from my experience with one special single game against a former
version of Crafty. I sent Bob the notation of the game. His main reaction was
not to be understood at the time: He wrote s.th. like the following: you must
play Crafty on a P 200 ...

And I had informed Bob about my observation that Crafty seemed to play worse
when I had played a move he didn't expect/foresee. Seems to be a principal
weakness in non-rapid games for Bob's feature. All this was discussed recently
in another thread. Could one win strength by playing mainly the limited (known)
search depth of the opponent? Seems to me that Bob did some interesting stuff.
Ok, it's mere clearvoyance again. :)

Enrique Irazoqui

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

"Christopher R. Dorr" <crd...@cinternet.net> wrote:
>Forfeit the match after one game? That's absurd. One game proves exactly
>nothing. It may be suggestive of something, but if you feel that this

>proves something one way or the other, I can recommend a number of good
>*introductory* books on statistics.

These books would tell you that a 10 games match is not much statistic
proof either. To begin with, statistics had nothing to do with this match.
Or else you would have to play 500 games. No, it had to do with how a
search handicap would affect the game, which you could analize in the game
itself. 10 games would have been better, and 1,000 games even better.
Still, statistics is by no means the only way to find out about things.

Enrique


Kenneth Sloan

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

In article <333A48...@pop1.sympatico.ca>,
Komputer Korner <b1bx...@pop1.sympatico.ca> wrote:
>Douglas Crowe wrote
>> ...

>> Excuse me for interupting,but is there a place where this
>> game between Crafty and Rebel can be followed,or is it
>> just updated here in the group?Thanks in advance.
>> ...
>
>http://www.xs4all.nl/~rebchess/

This is incorrect.
The correct URL is:http://www.xs4all.nl/~rebchess/match.htm

By the way - the information there still promises a 10-game match.

I was interested in the "calibration match" - until I downloaded the
first seven games and discovered that Rebel has played white in all
games.

Passing strange...


--
Kenneth Sloan sl...@cis.uab.edu
Computer and Information Sciences (205) 934-2213
University of Alabama at Birmingham FAX (205) 934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/info/faculty/sloan/

Robert Hyatt

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

Kenneth Sloan (sl...@cis.uab.edu) wrote:
: In article <333A48...@pop1.sympatico.ca>,

: Komputer Korner <b1bx...@pop1.sympatico.ca> wrote:
: >Douglas Crowe wrote
: >> ...
: >> Excuse me for interupting,but is there a place where this
: >> game between Crafty and Rebel can be followed,or is it
: >> just updated here in the group?Thanks in advance.
: >> ...
: >
: >http://www.xs4all.nl/~rebchess/

: This is incorrect.
: The correct URL is:http://www.xs4all.nl/~rebchess/match.htm

: By the way - the information there still promises a 10-game match.

: I was interested in the "calibration match" - until I downloaded the
: first seven games and discovered that Rebel has played white in all
: games.

: Passing strange...

Ed explained this. the auto232 doesn't alternate colors. He has to do
it manually. So typically a match will be N with white, then N with
black, making sure that the matche ends after M*N such series, where
M is even...

Bob


Roy Mongiovi

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

What I'd like to know is why everyone seems to invest so much EGO in
this match. It maybe makes sense for Ed, since he has financial
investment in Rebel, and for Bob, since he's trying to do a good thing
by making crafty available as a breadboard for computer chess
development.

But I'll be damned if I can figure out why the rest of you jump on
everything either of them says about their programs or the match.

If the match could have shed some light on the truth of tree searching
versus evaluation function, I think that would have been a good thing.
It seems to me that one game isn't enough.

Here's my hypothesis. Tree search is necessary to find combinations.
A chess program can't play a combination shorter than its tree search,
since it doesn't plan from move to move. In the absence of such a
combination, the evaluation function has to be able to discriminate
among all the positions it examines in order to guide the search toward
the most promising continuation.

So I guess the question is what percentage of the time can an excellent
evaluation function avoid positions where there are tactics beyond its
search capability, but within the search capability of a faster
program?
--
Roy J. Mongiovi System Support Specialist IV Information Technology
Georgia Institute of Technology
Tough are the souls that tread the knife's edge Atlanta, Georgia 30332-0715
Jethro Tull - "Passion Play" r...@prism.gatech.edu

Robert Hyatt

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:

Never said that. Don't know where you imagined it from, but it wasn't from
me. I clearly stated that after 10-12 moves, even without the book, crafty's
position looked perfectly sound to me. With a book it would have played
differently, but that doesn't mean a thing in this context. Because it
still could have come out of any book line with the wrong idea about what
is right and wrong. Maybe, maybe not. In any case, you won't find a single
post from me saying the game was lost due to a missing book, or the game was
lost due to a poor position that the book would have avoided. I'd take the
position at move 12-15 any day, for either side. Nothing looked bad to me
at all about the position there for either side.


: Look, that's what I criticised in a former post. If you argue like that you


: could wipe off all the brilliancy and magical of every single won game of
: Mikhail Tal for instance. You simple argue backwards having the later results
: in mind. And you *show* that after this or that Tal never had the chance to sac
: this or that.

: If you continue for some time on this -- you'll end in writing books about 1.e4
: and wins. Or 1.d4 and mate in 45 moves. Or you could say, that you end up with
: playing against god if you start playing chess. It's the same with the egg and
: the duck. Or what, if our parents wouldn't have had such a ball we all wouldn't
: be here ....


: For me at least Rebel won a nice game in a style Kasparov couldn't do better
: against Deep Blue. So we had in nucleus what will happen in May. I mean, it
: became quite clear that Rebel had enough chess wisdom more than Crafty to
: equalize/neutralize the exhorbitant time factor 100.

: Don't you agree on this?

in this one game, yes. More games would be more revealing. For example,
in the Korrespondence Kup, rebel and Crafty are using equal time, and the
position is dead even. So either Crafty is lucky to be alive in the Kup
game, or Rebel was somewhat lucky to survive with a huge search handicap
in the NPS game. But there's luck somewhere. Where, we can argue about
for a long time. I plan on discovering which is right myself, with data,
not discussion...


: Ok, and I'll agree that this single game didn't prove anything scientifically.

you might well be right. remember I've predicted 4-2 again for Kasparov.
2 wins, 4 draws. Just a wild guess. If Hsu and company have tuned the
eval over the past year, this could *easily* change for the better of
DB. We'll see in a month or so...


: >He has said numerous times that he is not tuning the engine AT ALL. He is


: >attempting to rectify problems with the autoplay interface.

: >Why are these things so hard to understand?

: Because it's not true. If I had the time I could quote Bob's writing about
: several versions in the next 6 months. And he surely wanted to *tune*. The
: book, the passed pawn stuff and so on. You must read a bit more carefully. But
: I agree Bob did never propose to enter version 11.19 or 20 for game 2 or 3. But
: all the wining meant exactly the same (implications wise ...). Let me even go
: further to baite Bob. He dreamed of a tuned version which had learned enough
: about Rebel to master his slave thereafter. And I'm even sure that this would
: be possible to implantate.

Nope, not even close. there was no whining. I simply reported here what was
going on with the match, and why there were new versions every day, and what
these new versions were changing. Never changed the eval. Except of course
that fixing the timing bug effectively changed the eval quite a bit because
the gross contempt score was no longer non-zero due to a timing mistake.
Never changed the search. Only changes were to the auto232 interface, which
is still not working as it should...

: Bob, I'm not doing anything spooky. It's only exact (!) reading of your


: writings. (:-))={ I don't know, but I always hit the ceiling these days.
: What's this all about? I can really fly. Me thinks.

Then you ought to quote what you read. I talk about new versions of Crafty
here all the time. Ed and I also discussed this point at length and we
both agreed that we'd stick with the same programs throughout, other than
we each could fix problems that were unrelated to the chess code, such as
the timing problem both he and I had, as well as the auto232 problems I
had. There was never any attempt to do anything but let the two programs
play to the best of their abilities, with us taking care of any odd thing
that came up. In a real match, for example, Ed couldn't tinker with the
time per move once a game started. We both agreed he not only should be
able to do so, but that he should be required to do so, so that Rebel
would not suddenly start moving too quickly. That was *not* in the spirit
of what we wanted to show. Therefore we agreed that he could adjust the
time per move whenever he wanted, since the code that normally did this was
not being used due to the odd circumstances of the match with no thinking on
the opponent's time allowed.

You read, you misunderstand, then you misquote. I know you can do better...


: >Chris

: Greetings and thanks

: Rolf

: P.S. Remember I gave my betting at the beginning of the game with 6 to 7 points
: FOR Rebel as the winner. Maybe that was for a great joy all over this planet
: ..

: But I did this from my experience with one special single game against a former
: version of Crafty. I sent Bob the notation of the game. His main reaction was
: not to be understood at the time: He wrote s.th. like the following: you must
: play Crafty on a P 200 ...

I believe my response was "play the real crafty" on icc/chess.net. I have no
idea what version you used, what book you used, what settings you used, etc.
Doesn't take much to make it screw up badly. The one on ICC doesn't...

: And I had informed Bob about my observation that Crafty seemed to play worse


: when I had played a move he didn't expect/foresee. Seems to be a principal
: weakness in non-rapid games for Bob's feature. All this was discussed recently
: in another thread. Could one win strength by playing mainly the limited (known)
: search depth of the opponent? Seems to me that Bob did some interesting stuff.
: Ok, it's mere clearvoyance again. :)

and I don't know what this might mean. If you'd like to play a game, here on ICC,
where it is *forbidden* for you to *ever* play the move Crafty expects... I'll be
happy to oblige, and will put money on that outcome. (hint... any piece defended
only once is not defended at all should crafty take it... :) )


Preston Briggs

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

Well, I'm very disappointed to seem the match terminated.
Been enjoying the game. Indeed, about 10% of my company has
been following the game, stopping by my office every day
to check the progress.

It's too bad, after all the initial work and the time invested in the
play, to see the experiment trashed without result. Some people may
think that the first game is significant -- I disagree. I'd say the
first game was more along the lines of a shakedown run. It's
unfortunate that provisions weren't made for such a test game to
ensure that all the experimental apparatus was in place and stable.

When you do experimental work like this, you have to be prepared to
redo your experiments. Happens in lots of circumstances

you have to be able to reproduce your results

others have to be able to reproduce your results

something went wrong and you have to correct it and restart
the beginning

These things happen all the time. It's why science, especially good
science, is expensive. I can't tell you how many time I reran all the
experiments described in my thesis. Over and over and over again.
Very depressing. But in the end, the results were good (meaning well
documented and reproducible).

I encourage you guys to reconsider the experiment. It seems clear
that a lot of people are interested in the result. Indeed, even the
problems you have uncovered have been interesting in their own right.

Preston Briggs


Enrique Irazoqui

unread,
Mar 27, 1997, 3:00:00 AM3/27/97
to

Roy Mongiovi <r...@prism.gatech.edu> escribió en artículo
<5hega3$4...@acmex.gatech.edu>...

> But I'll be damned if I can figure out why the rest of you jump on
> everything either of them says about their programs or the match.

For the same reason you just did. It's an interesting topic, no?

Enrique


Komputer Korner

unread,
Mar 28, 1997, 3:00:00 AM3/28/97
to

Roy Mongiovi wrote:
>
> What I'd like to know is why everyone seems to invest so much EGO in
> this match. It maybe makes sense for Ed, since he has financial
> investment in Rebel, and for Bob, since he's trying to do a good thing
> by making crafty available as a breadboard for computer chess
> development.
>
> But I'll be damned if I can figure out why the rest of you jump on
> everything either of them says about their programs or the match.
>
> If the match could have shed some light on the truth of tree searching
> versus evaluation function, I think that would have been a good thing.
> It seems to me that one game isn't enough.
>
> Here's my hypothesis. Tree search is necessary to find combinations.
> A chess program can't play a combination shorter than its tree search,
> since it doesn't plan from move to move. In the absence of such a
> combination, the evaluation function has to be able to discriminate
> among all the positions it examines in order to guide the search toward
> the most promising continuation.
>
> So I guess the question is what percentage of the time can an excellent
> evaluation function avoid positions where there are tactics beyond its
> search capability, but within the search capability of a faster
> program?
> --
> Roy J. Mongiovi System Support Specialist IV Information Technology
> Georgia Institute of Technology
> Tough are the souls that tread the knife's edge Atlanta, Georgia 30332-0715
> Jethro Tull - "Passion Play" r...@prism.gatech.edu

Yup, you summed up the question quite well, but how do you vote?
Deep Searcher or Knowledge King?
--
Komputer Korner

The inkompetent komputer.

Roy Mongiovi

unread,
Mar 28, 1997, 3:00:00 AM3/28/97
to

In article <333B82...@netcom.ca>,

Komputer Korner <kor...@netcom.ca> wrote:
>Yup, you summed up the question quite well, but how do you vote?
>Deep Searcher or Knowledge King?

Well, I'd have to say I vote for the "Knowledge King", provided you get
a reasonable search depth (which I think is pretty easy given the speed
of today's processors).

I think an evaluation function can approximate a deeper tree search
(with an exchange evaluator), and give the computer a better chance of
playing good moves if it has lots of smarts. Tree search just isn't
deep enough to find tactics in all positions. If there isn't an
amazing tactic, you need an evaluation function that can distinguish
between really good positions, and not so good positions.

I think that if the evaluation function isn't smart enough, there are
lots of effectively "do nothing" moves, and it makes it easier for the
program to insert do-nothings and exacerbate the horizon effect.

If the evaluation function can't tell two positions apart, tree search
doesn't help.

Of course, what I really want to know is when does a little version of
Deep Blue go on the market :-)

mclane

unread,
Mar 28, 1997, 3:00:00 AM3/28/97
to

"Torstein Hall" <tors...@eunet.no> wrote:


>mclane <mcl...@prima.ruhr.de> skrev i artikkelen
><E7pKz...@news.prima.ruhr.de>...

>> "Christopher R. Dorr" <crd...@cinternet.net> wrote:
>>
>>

>> Statistic proves nothing ! A chess program consists of positions. Each
>> position can be evaluated. The programs give an evaluation, and they
>> search. Looking on each evaluation and on each main line gives you (in
>> a normal game of 50 moves) 100 events where you can SEE and this way
>> proof if a program plays good chess or not.

>Why are you always wrong? :)

I am not always wrong. I have said Rebel8 will be the leader of the
ssdf-list long time before they tested it, long time before any
error-margin could have shown this and with only a few samples of
games.

And I can tell you more examples like this. But thats all not
important. Or ?!

Sorry , there is an error in my above statement. I wanted to say:
A chess game consists of positions.

>Statistics prove a lot, within certain limits.
>Since chess consists out a very big number of positions, its impractical to
>test them all! You rather play some games, study the eval of the programs,
>and the results of the game and make your conclusion about most of the
>positions.
>Then you say with that one program is stronger than the other, within your
>margin of error. If you can calculate how big the probability of you beeing
>right or wrong is, then its statistics!


But statistics does not proof. It can only give numbers. And you
cannot say much about the event what cannot be said WITHOUT
statistics.

Anything we know from life, can be found out WITHOUT statistics.
Statistics is the inkompetent post-game-analysis of people who think
they have to explain what is clear to anybody.


mclane

unread,
Mar 28, 1997, 3:00:00 AM3/28/97
to

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:


>There's lots of doubt, Crafty's played stobor many times on both ICC and
>Chess.net... log on to ICC and "search +=crafty -=ratbert" and then
>"search -=crafty +=ratbert"

>The first command produces 65 wins for crafty... the second produces 10
>losses for crafty...

>As the saying goes,

>Q.E.D.

I do not believe in other people's analysis much.

I have stobor. I have crafty. I let the play against each other and
study the games.

Shall I show you the games in EMAILS ???

>: I don't understand why now crafty killed stobor. I though it was vice
>: versa.

>you sent me 3 games. the first, "stobor kills crafty, *but* stobor made
>a mistake in the endgame and lost." the second "stobor kills crafty, *but*
>stobor didn't realize that it had to stop the a-pawn from advancing and it
>lost." the third "stobor kills crafty, but crafty was lucky and found a
>repetition draw."

Right. THIS Stobor was BUGGY , and I thought Tom will fix these BUGS
before he participates in Paderborn. As he has shown, he participated
without fixing them. What a pity.


>Those were what I was talking about. It sounds like the old elementary
>school conversation, where a kid comes home with a black eye, blood all
>over his shirt. His dad asked "did you get into a fight?" The kid
>responds, eyes down, "yes.". Dad says "how'd you do?" Kid responds,
>"I beat the hell out of his fist with my face." :)


Come on. Why can't you accept that THE RESULT is not always that much
important like the content of the game.
I mean: would you mary a girl only because you have heard about her
size, weight and the color of her eyes without meeting her ?
Is the RESULT, the NUMBERS anything to you , guys ?

If you wish I play more games. But I stopped it because TOM said he
has made a new version , and I thought, my versions are too old now.


>: >Oh god no. That was Thorsten, not Tueschen. Let's not go down that road again.

>: >bruce

>: Do you really believe that I would have posted stobor kills crafty if
>: I would not be sure about this ?!
>: I have both programs, I have machines to find it out.
>: Nothing is easier. You don't need 100 or 200 games to show evidence.
>: If you have eyes to see and feelings a small sample says more to you
>: than any autoplayer can find out in 50 or more games.

>Exactly. I have enough evidence to convince myself.

Hm. There is really a big sea between you and me.
Not only the distance of different age, but also the different
point-of-view and ideology. Strange.

>: I don't understand why you have problems to believe that feelings can
>: replace statistic perfectly. Don't you love your wife although never
>: any evidence and statistic was made about your love ?!

>: BTW: The poster was mclane !

>I corrected it and appologized after Bruce pointed out the error. :)


No - I wanted to say: please use mclane instead of TH....

mclane

unread,
Mar 28, 1997, 3:00:00 AM3/28/97
to

pre...@cs.rice.edu (Preston Briggs) wrote:

>Well, I'm very disappointed to seem the match terminated.
>Been enjoying the game. Indeed, about 10% of my company has
>been following the game, stopping by my office every day
>to check the progress.

WHAAT ? 10 % of your company followed the game ?!
Unbelievable. WHY was this so interesting for you ?!


>It's too bad, after all the initial work and the time invested in the
>play, to see the experiment trashed without result. Some people may
>think that the first game is significant -- I disagree. I'd say the
>first game was more along the lines of a shakedown run. It's
>unfortunate that provisions weren't made for such a test game to
>ensure that all the experimental apparatus was in place and stable.

>When you do experimental work like this, you have to be prepared to
>redo your experiments. Happens in lots of circumstances

> you have to be able to reproduce your results

> others have to be able to reproduce your results

> something went wrong and you have to correct it and restart
> the beginning

>These things happen all the time. It's why science, especially good
>science, is expensive. I can't tell you how many time I reran all the
>experiments described in my thesis. Over and over and over again.
>Very depressing. But in the end, the results were good (meaning well
>documented and reproducible).

This is my job too. But I have to find out BEFORE scientific-statistic
proofs my ideas.

Robert Hyatt

unread,
Mar 28, 1997, 3:00:00 AM3/28/97