Author: Vincent Diepeveen
Date: 07:06:39 08/29/00
Hello computerchess fanatics,
Here the huge impression i had with DIEP at the WMCC.
Previous WCC i had a bit of comment from Ernst A. Heinz, but
also an overwhelming amount of positive reactions at a description
of the WCC from my viewpoint.
Especially seeing what the shortcomings were is very interesting for
the future. That it looks like a big excuse for what is wrong, as
dr. Ernst A. Heinz suggested after previous report in 1999
sure isn't the truth. Without analysis you
can't continue. In fact how many THINGS have you read about the
WMCC yet? Probably nothing except the results of the games and
some people who still call shredder 'lucky' to win the world championship
tournament for the 3rd time. Without seeing things that are wrong or went
wrong you can't improve anyway. Note that what is in here is MY VIEWPOINT.
It is NOT the viewpoint of any organisation or company. If people disagree
and quote, please let them not quote a single line, but the whole picture;
the reason for that is quite obvious: i'm born in a country where we don't
speak much english, so misformulations of a single line should be taken
with a bit of salt. It's the whole picture that makes the story!
Note that the basic difference between Heinz and me seems to be that i can play
some chess, about 2254 points higher rated and therefore i realize very clearly
what's happening in the computer chess world, where Heinz still says that it
is not proven that knowledge works!
Here another story how a bit of far endgame knowledge won again the title for
Stefan Meyer Kahlen, apart from some hotel complaints and things
about my own program DIEP.
This World Computer Chess Championship was held in North London in the
Alexandra Palace. Let's not talk too much about before the WMCC. Basically
this has been discussed extensively everywhere. I'll limit to a few lines.
- there were no announcements nor any other information about the
tournament. Also despite repetitive questions to Don who seemed to
be in charge of the 'organisation' of the tournament, instead of Levy,
i never received any email regarding any planning.
I did get an email about the cost of hiring a monitor. Instead of
freely giving one to everyone as in WMCCs/WCCs in different countries as
UK, they asked instead 400US$ just for hiring a 15'' LCD monitor.
All i need is a SVGA monitor. Lucky this was exchanged. We would get a 17''
XGA HP monitor for 40 pounds. Now i find this also quite a high fee, but i
accepted
it, as the alternative was to carry one myself.
I got onto the train with a lot of problems
- my laptop had broken
- my nino is either stolen or has vanished somehow
- diep crashed
In total i had slept 2 hours before getting onto the train. Especially
the fact that i simply had not prepared a good tournament book and that DIEP
crashed i didn't like.
I met Franck Zibi (zchess) in Paris at Saturday 19 august. Lucky Franck arranged
the trip with the Eurostar to London.
Now a lot of people are a bit paranoia in the computerchess world, but the
French
& British government are a lot worse as that.
After being in a row for over 40 minutes and a double check on the ticket and
passport,
all hundreds of people who were waiting for the eurostar were driven together as
cattle and then a few police officers with dogs ran through the crowd
suspiciously
looking to everyone, and especially letting their drug dogs smell at the phones
for
drugs. Sitting on my computer one of the dogs nearly jumped over me. Now lucky i
don't fear german shepherd dogs, which jumping over the luggage as there was
simply no place to stand over there. Hopefully a lot of small kids&babies aren't
scared to
death for life.
After all these searches some more groups of japanese (lots of japanese in
Paris!)
were driven into the small area where we had to wait, together with a lot of
other
'fresh' tourists, as the eurostar had another hour delay.
After that we went through another few ticket checks, before we were on the
eurostar.
Again i noticed that those trains drive faster as nowadays the infrastructure
allows,
especially the thalys which drives with 300 KM/hour between Brussels and Paris
was resonating in a way which is not very happy. A single problem will directly
leave hundreds of people dead! Amazing so little accidents happen with trains
at this speed! Perhaps because they get driven by professional drivers who
do a way better job as any artificial intelligence currently is doing?
In the evening we arrived in London. We took a cab to our hotel and freshed up
there. Now we paid in advance for the hotel, a quite silly thing to do as we
found out. 290 pounds we paid for a double room, including breakfast for 6
nights.
Now hotels in London are usually located in old buildings. Whole North London
made
a quite outdated impression at me, but take into account that the Alexandra
palace
area is a very small part of London. The power at the bathroom didn't work, as a
result of that i had an excellent excuse to explain why i was unshaved during
the
whole tournament. The shower didn't work, lucky there was also a bath which
didn't
leak too much, the towels got never changed and the room was only cleaned
a few times in the week we stayed there. In netherlands you get this for
1/3 of this price, but then with clean towels and with a cleaned room.
Only in the hotel i discovered why i was sweatting that much. My luggage all
together was 60 kilos!
Sunday Morning, 1 day before the tournament started, we went to the Alexandra
Palace.
We walked down the hill and then up the hill using the normal road that went
with
a lot of turns to the alexandra palace. Lucky we didn't bring the computers yet,
as we would have nearly died doing so.
To my amazement the palace was not only a real palace, but also having real huge
halls.
Also it clearly was built on top of a big hill.
In a huge hall about 900 kids were playing chess. The corridor
was used for all other kind of games, like stratego. Later i would find out
that there is still a lot of space to expand the MSO games as another even
bigger hall was completely empty.
I realized why nothing had been done on the organization of the computerchess.
Levy
had been more as busy organizing all these event!
Unlucky in this hall we were planned to play as well. Also there were no
monitors
yet, so there was no chance to fix all the problems in the different programs.
We
were forced however to directly pay to the tournament organization the fees.
As they didn't have small coins to return to me, i paid 67 pounds entry fee
instead
of 66.67.
I was very dissappointed by this, as i had so many things to fix. Especially the
fact that my win32 interface wasn't working yet caused that i had done nothing
yet on the tournament book. Also i was having a bad feeling about the crashing
of
DIEP short before WMCC. The last thing i wanted to do was losing games because
of
multiple crashes.
Having nothing to do i gave a simultaneously exhibition on a few boards without
clocks. The first participant of the simultaneously exhibition was Franck, the
second surprise surprise: a chess program. Called Pacque Expert.
Franck missed some chances against me, but Pacque Expert lost chanceless in the
special Diepeveen-anti computer attack (1.e4,e5 2.nf3,nc6 3.c4?,Nf6 4.Nc3,Bc5
5.h3
followed by g3 d3 Bg2 O-O and then win the game based upon f4-f5 and g4-g5).
Now we were told by Don Beal that we could setup after 6 PM, when the kids would
have finished playing.
In the afternoon we ate in a pub/restaurant. The itialian spaghetti bolognese
tasted real
good to me there. We would come back quite regurarly at this pub/restaurant.
Amazed i
was by the high prices for drinks. Even coca cola was real expensive.
Now we took the shortcut to the alexandra palace, as described by Andrew
Williams.
We were real lucky with this remark from Andrew, as carrying each a
computer+equipment
is not our favourite sport.
Arriving in the tournament hall it appeared that the kids were still getting
awarded prices and partly still were playing.
Only 2 hours later we could help setup where we would play.
Regrettably there were no monitors at all, except a monitor from Don Beal
himself.
So at about 8 PM i was allowed to test first. My computer didn't work
regrettably.
We lost a lot of time there. After all it appeared that my computer only worked
when it had 2 cpu's. I had taken out a cpu in the morning and putted it back.
After this my computer worked again. I find it quite weird that the supermicro
motherboard is this sensitive. In the manual it says about the P6DBE/DGE that
it doesn't matter whether there are 2 inside or 1. There is no mention of
anything like a terminating card etcetera.
Also the bios is a big mess. The latest update of this bios allows slot1 800Mhz
cpu's and up to run at this motherboard (but at maximum 100Mhz bus, even though
the BX chipset allows overclocking without problems and i have sdram which
will go up to 133Mhz SDRAM inside the computer).
Buying a slot1 motherboard (except for the asus P3C-D with the expensive RIMMS)
seems not a very smart move nowadays, even though i personally
find the idea of slot1 cardridges
a lot nicer as fiddling with cpu's where all kind of pins can break.
When i got back in the hotel i basically concluded that except for a good meal
in the afternoon i had done nothing interesting the whole day and basically lost
a full day.
Monday morning when we got to the tournament hall, there were some problems
again.
First of all we were told the day before we could get in at 9 AM. However at 9
PM
the security didn't allow anyone in.
When v/d Herik and Don Beal walked as 'organizers' into the tournament hall, i
followed
them. The rest of the programmers would need to wait for over an hour outside
before they were allowed in by the security (even though we all had buttons).
When i unpacked the fresly arrived 'new' monitors, i directly recognized these
were old samsung 17'' monitors, instead of the promised new 17'' XGA HPs.
With another 40 pounds more i'm sure that Christophe Theron with some shopping
down
town could have found one for this price (assuming he would want to carry it
many miles up hill).
I didn't find the bug which caused the crash in DIEP, but i was quite sure the
bug
was caused by the EGTB probing code in the qsearch, so i turned this off.
For the rest of the tournament diep would only probe again in the mainsearch
instead
also in the qsearch. IMHO probing in the qsearch is necessary for any chess
program
to be sure it works 100% cool, however it's in 99% of all cases not influencing
results.
Regrettably i again hardly had time to prepare something for the first game.
When the security gave clearance to allow everyone in the playing hall, it
appeared
that many also had huge problems carrying equipment to the tournament hall.
Playing a computerchess tournament in a surrounding dominated by hills and
valleys
is simply not a good idea unless you go there by car.
When the first round started DIEP played as prepared the Sveshnikov with black
against Shredder. When Shredder played the classical mainline i could tear off
my
hair, as i could have known this. Diep was not quickly out of book and left book
with a -0.00 score (being a score between -0.001 and -0.009). Yet it played the
very bad Kh8 instead of Ra8-a7 as my newly bought books described. So DIEP was
strategically chosing the wrong plan and within a few moves score jumped to
-0.80
During the game i was amazed that shredder at over 2 times faster hardware as
it had at Paderborn was searching seemingly less deeply as it did in Paderborn.
Also the variations were a lot longer. Seemingly Stefan decided to be less
selective
and be tactical quit a bit stronger. Note that these longer variations also
might
get explained by doing more probes in the hashtable. In DIEP i'm used to
get in the endgame at search depths of like 12 ply variations up to 20 unforced
ply, no problem.
Now obviously DIEP was lost and i started realizing why no other commercial
book of a chessprogram plays Sveshnikov-Pelikan.
Sveshnikov as a modern opening is more than ok, but it's heavily
strategical dominated when playing it with black.
Not chosing the right plan with black is simply desastreous against white being
a computer, which is actually what happened short after DIEP was out of book.
Nevertheless i hope to continue playing Sveshnikov with DIEP now and then.
Chosing
the right plan with black is a matter of a bit more tuning and a bit more
knowledge,
though this will take quite some time. Strategics is hard to do right.
Lucky DIEP after this bad start attacked quite good and managed to exchange to a
dead drawn position. Here Shredder really evaluated the opposite bishops clearly
much
better as DIEP. Even though DIEP knows opposite bishops it had a very
fluctuating
score from -0.98 to -1.7. Shredder had about +0.6 to 0.7 for itself, a much
better tuning of the opposite bishops obviously. Because of the bad tuning in
DIEP
it exchanged the bad bishops to get into a rook endgame. Regrettably this
rook ending was lost very simple. Shredder had no mercy despite some tough
opposition of DIEP.
After the game i figured out that i just had to tune the opposite bishops 0.1
pawn
more penalty to already not play the extremely losing move that got diep to the
lost rook endgame. Also improving rook endgame obviously will pay. Where i
improved
the complex middlegame drastically after the WCC99, obviously the far endgame
is the next thing to do in DIEP.
Note that a bit better book would have helped considerable also, though we must
not forget that DIEP actually OUTPLAYED shredder in the middlegame, after both
realized the strategic problem of black.
As this game took so long i didn't have time to prepare for the second round
either.
Crafty lost amazingly simple to Rebel. Really amazing because crafty lost
with exactly the same line in the 1996 WMCC in Jakarta.
EXACTLY the same line. Crafty went for a rook, got the rook and
a few moves later resigned. It's not my business, but crafty is gonna be
forever chanceless in a serious tournament if it continues to lose games in
the first few rounds in such a silly way.
The second round DIEP was paired against Junior. Before the tournament started
i had figured out that it was perhaps a good idea to play 1.e4 with DIEP. I
stuck to this idea, turned on 1.e4 and started playing against Junior. Note that
i had prepared one small line against junior: Najdorf.
Regrettably this opening didn't get on the board, but Richter Rauzer instead.
Richter rauzer is a very modern opening which i play with white too. It's a very
passive opening for black initially. I was not amazed that after Juniors 12..Ra7
DIEP was out of book. After DIEP's answer Qe3, which is a real good move
seemingly,
Junior was out of book.
Now DIEP didn't play like an Anand here, but sure didn't do very bad. Yet after
18.e5
i had a huge fail low (big drop in score) in opponents time for dxe5 as black
can
pin the white knight on d4 with an early Bc5.
Lucky Junior missed this golden shot after which DIEP found the great shot
21.Qg5
e3-g5 (1:31:20,1:20:41) 1.25 11 7820808
01:51 4174445 (0) 10 1.25 Qe3-g5 Ke8-f8 Nd4-f3 d6-d5 g2-g4 h5xg4 Re1-g1 Bg7-h6
Qg5xg4 Kf8-e8
Note that Junior had seen Qg5 a move sooner already as it played 19..h5 and not
the
expected 19..Bg7 because it had a fail low after 19..Bg7? 20.Qg5 and then after
Bxe5
Rxe5!
Well seen by Junior. Basically this tactical finishing touch is what DIEP lacked
in this game, as score got higher and higher, but DIEP didn't manage to convert
it
into an easy to win position. Amazing but true is that Junior hardly evaluated
the
position better as white. Only when it was for everyone clear that white was won
bigtime, junior gave white a slight plus then. The rest of the tournament one
would see this phenomena. Against Nimzo junior even thought it had a better
position
(junior initially a pawn down), till short before it lost.
Regrettably DIEP exchanged pretty late into a rook
endgame in the second round against junior
which was very tough to play. A lot of times i had a drawing score,
but after having done both a draw offer, DIEP score went up again to 0.80 short
before the first time control.
Obviously it was gonna get a long game. About 15 minutes before the time control
at move 60 i adjusted time.
Now we were near the first time control and with 6 minutes left on clock DIEP
had
just 1 move to go. I was so asleep that i hadn't noticed that DIEP had left on
its
clock only 5 minutes and a few seconds where the internal clock of DIEP said
5 minutes and 53 seconds.
The small stupid clocks we played at, it was seemingly having 6.30 minutes left
on clock.
Also I had completely forgotten also that nowadays winboard versions of DIEP
don't
leave any time left on their clock if they fail low.
They can use up the completely left amount of time
on clock. In past times this couldn't happen. DIEP would always leave a couple
of
minutes on clock left. When not getting a fail low, diep would have played its
move within 2 minutes here. So even with less time on my clock it normally would
be ok.
Yet the first disaster had happened. In a position which objectively was a draw,
but where Junior was doing shit, DIEP managed to forfeit as it got a fail low
which it couldn't resolve.
I guess DIEP would have had chances to win this game, as Junior lately was
clearly
not actively exchanging to a draw.
Just half a minute it flagged DIEP made its move as its internal clock had
reached
0:00.00 left till move 60.
So i started with 0 out of 2. Not a very happy start, and seeing many programs
with the same interface i feared even worse things to happen.
Now the 3d round i played Pacque Expert. Normally this program should not be
allowed to play a world championship as semi-professional.
The ICCA dudes even forced him to pay 250 dollar
entry fee and he was sitting in the Raglan hotel as adviced by the ICCA. A very
expensive hotel. Over a 100 dollar a night single room or something.
Pacque Expert was mated real soon even though diep played a bad openings line
with
black. Pacque Expert gives away material, even when searching 12 ply in assembly
(which is 16 bits and dated 1990 by the way).
In this case it gave away a bishop against a single passed pawn (though
connected)
and in total 2 pawns.
Pacque Expert is a very weak program, and DIEP is a very human playing program,
so i'm not sure of this, but surely DIEP finished pacque expert real quick. It
was
mated within a few moves.
The 4th round (second round that tuesday) DIEP had 1 out of 3 and faced Zchess.
Now i knew in advance that some disasters might happen with diep with regard to
book and passed pawns, but i sure didn't realize that it would already happen
that early in the game.
Basically DIEP played the strategical awfull c4 against Zchess because of a bad
passed pawn evaluation. DIEP sees already when it plays c4 that black can get a
passer at c3. Basically this determines the game strategical in the advantage of
black.
Also very bad is the move Re4-h4. This, though real simple for mankind, is hard
for
programs to see, as the rook is not hung at all. It's just CLOSED in, but it
continues
to put pressure at h7. So the piece is not hung at all, nor trapped or anything
of that kind. It's 'just' out of play forever.
A few moves later i resigned. I was not happy with DIEP performance here and i
sure plan to bugfix the passed pawn code as well as the Rh4 move real soon. As
Rh4
is just a matter of a bit of better knowledge in the closed in code.
Round 5 at Wednesday morning i played Insomniac. In dutch open diep couldn't
finish the endgame against insomniac after being dead won. In CCT i lost from it
after a bad opening as well as pathetic play of DIEP.
This tournament it would go different. I had selected now the taimanov in the
sicilian. I really find this a real cool variation, but regrettably i had
prepared the wrong lines to play against insomniac. Insomniac took care diep
transposed into the Kann. The line played is something i play myself also
with white, and it plays real easy with white, let's face the truth.
Basically the a6 move and b5 concept sucks bigtime against a computer. White
gets left with 2 half open files. Just putting all pieces to the center is
enough
to win at least a pawn and/or the game for white then.
This is exactly what happened. Diep played h5, a move it has played thousands
of times before in blitz games. It seemingly played it without good reason.
There must be really a small detail wrong, unless we see h5 as a kind of
nullmove that's performed in order to avoid all kind of tough probs in the
center. I feel that searching 10 plies here didn't do much good either. Several
times i saw a fail high to a good move which wasn't finished in time.
Here i really felt that my book preparement was just 1 week too short. With
several
full days more i sure would have avoided the transposition to the Kann, or
forced
it to play a better variation as the b5? line in the Kann.
The interesting thing is that this line is very popular in human games, so the
fact that the wide book selected this line was very logical. This
openingsvariation
is a classical example of something which a human can play but a computer can't.
The only program i could imagine which might have been able to play this
with black is crafty, as it really likes pawns on d6 and e6.
Diep at least didn't like this position at all. Already the nullmove h5 is
played with a score below zero, though quite small. Insomniac itself
thought it got out of book with about 0.5 to 0.6. When the probs occured
in the center, then DIEP went down from -0.18 to -1.0 within a move.
Weirdly insomniac only very late realized it won a pawn. As James already feared
it was not very smart to exchange queens. The Bh4 move was a fail high DIEP had
at quite a big depth. It's a superior move. Diep expected here already that it
wouldn't lose a pawn at all, but 'only' face a strategical bad position. When
Insomniac decided to win the pawn i already had a 0.0x score.
However to my big surprise. Actually also to my dissappointment the resulting
position wasn't won by DIEP. I thought personally that black could win this
easily by just exchanging on a5 and then winning with the b pawn. Tactical this
wasn't so easy however. Some last minute fail highs from Insomniac took care it
easily defended the position. The game was given a draw when both programs
wanted
to draw the position by repetition. My score had dropped then from 0.08 to the
expectation that it would be a draw by repetition valued 0.000
Afterwards i couldn't complain of course that DIEP drew this game. Getting lost
quickly then getting back in a dangerous endgame and drawing it after allowing
white to make some mistakes is usual not a bad performance.
Only the book maker of DIEP should get some big punishment for allowing
transposition
to the Kann with b5. This is however not very hard to do. Just a small
investment
of time in the future and basically playing a bunch of games.
Later in the tournament i would get really impressed by how well tested
Insomniac is.
It won from Nimzo!
A great achievement, especially when you see the dead lost openingsposition from
Insomniac. Surviving such a position is real great. Now i'll not cry victory too
loud about Insomniac, because realism tells us that the achievement is not only
James part, but also the books part.
Basically the insomniac book is quite normal, just the books of some programs
like
crafty and DIEP suck incredible, even though there is great hope for those
programs in the future.
Very clear is that insomniacs achievements this tournament were real good.
Taking
into account the age from James this makes the achievement even better.
Personally
i question however how a bitboard program like Insomniac can improve its
performance.
It's just real good now, how does it ever get better?
What crafty is doing currently with bitboards seems to me the maximum achievable
with bitboards. After that bitboards look to me as something rather complex to
improve
evaluation and search with.
Anyway, if a bitboard program gets better as crafty, then it sure will be
Insomniac!
Right now i had the impression that good testing at the internet and the good
bookjob
from the friendly austrian was the main butt kicking reason.
In the afternoon it was time for blitz. Lucky there were still quite some
entries
for the blitz. In blitz clearly a few programs were real good and the rest
sucked in my eyes bigtime. Insomniac was a real positive surprise in the blitz.
It directly gets 10 ply in blitz, together with the fast operating from James
and the reasonable book this took care it could connect itself to the blitz top.
I found it too bad that SOS didn't join the blitz. I really find it a
good blitz program!
We really missed Ferret here, as now chessbase dominated clearly with the Kure
book
in blitz. Amazingly junior which i also find a real good blitz program did end
3d instead of first or second. Now there was a tie between Fritz and Nimzo.
Before this tie took place there was quite some time left to talk.
Fritz is the core program from chessbase. Nimzo is just another engine
they sell in chessbase eyes.
Of course there were team orders now. I can imagine that if i was owner of
chessbase that i would give the order that Fritz had to win the blitz. Seemingly
this is what happened; very CLEARLY before the game Kure turned of several
things
before starting the blitz games, even though Alex managed to do this within
a few seconds. As a result some incredible pathetic games were
played. I think everyone can see the bad games. When i asked Frans why Kure
had done this he gave the
official explanation: "We didn't want to show the tournament book".
Now i find this explanation sucks bigtime, as when i would operate my
engine in such a 'team' i would still turn on the tournament book in
Fritz.
Fritz is quite known within Europe, Nimzo hardly. It's obvious that chessbase
can
earn more when Fritz wins as when Nimzo wins. It's not clear to me whether
they for sure gave fritz the title, or whether they only increased the chances
that nimzo would not win the title.
Note that it's possible that Matthias Feist who operated Fritz
had missed the explanation of Frans, as i didn't see Matthias reset parameters
after the 'newgame' like Alex did. Of course it's possible that Matthias is
even faster with the mouse as Alex and resetted it when i blinked with my
eyes.
We can only guess what would have been the chances of Nimzo to win the
blitz title if there would have not been team orders. I would have betted
on Nimzo. For sure is that even if the explanation of Frans
is true, that it favours fritz, as Nimzo is really very tuned to
the tournament book.
The second game i didn't even watch
the end of it as i already knew the result in advance.
Anyway, the games speak for itself. Nimzo missing the queen win
in the first game is obvious enough showing that this wasn't the
normal book.
A different case were the blitz games before that. See DIEP-Fritz,
where fritz obviously is using its tournament book.
Note that DIEP played not bad, considering that there were a few 7 ply searches
at critical moments. I really regret DIEP allowed the passed pawn to run
further in the endgame. I figured out that at 8 ply DIEP would have played Bd6
preventing all fun for black. Diep would have won the game then most likely.
Note that i was amazed that just a few patzer moves of DIEP could lead to
a good position anyway. The blitz games were 7 minutes all by the way. As you
see i had to set a bit faster level as you lose some time when you operate
the program.
e2-e4 (0:06:00,0:06:00) book
c7-c5 (0:06:00,0:05:54)
g1-f3 (0:06:00,0:05:54) book
d7-d6 (0:06:00,0:05:49)
d2-d4 (0:06:00,0:05:49) book
c5xd4 (0:06:00,0:05:45)
f3xd4 (0:05:59,0:05:45) book
g8-f6 (0:05:59,0:05:40)
b1-c3 (0:05:59,0:05:40) book
a7-a6 (0:05:59,0:05:36)
c1-e3 (0:05:59,0:05:36) book
e7-e5 (0:05:59,0:05:31)
d4-b3 (0:05:59,0:05:31) book
c8-e6 (0:05:59,0:05:26)
f2-f3 (0:05:59,0:05:26) book
b8-d7 (0:05:59,0:05:20)
g2-g4 (0:05:59,0:05:20) book
b7-b5 (0:05:59,0:05:15)
g4-g5 (0:05:39,0:05:15) 0.16 9 752532 00:19 727625 (0) 9 0.16 g4-g5 Nf6-h5
Nc3-d5 Be6xd5 Qd1xd5 Ra8-c8 Bf1-d3 Nh5-f4 Be3xf4 e5xf4
b5-b4 (0:05:39,0:05:10)
c3-d5 (0:05:27,0:05:10) 0.07 9 471670 00:11 451299 (0) 9 0.06 Nc3-d5 Nf6xd5
e4xd5 Be6-f5 Bf1-d3 Bf5-h3 Qd1-d2 Nd7-c5 Qd2xb4 Nc5xd3 c2xd3
f6xd5 (0:05:27,0:05:03)
e4xd5 (0:05:22,0:05:03) 0.11 10 405947 00:03 144961 (0) 9 0.11 e4xd5 Be6-f5
Bf1-d3 Bf5xd3 Qd1xd3 Bf8-e7 Rh1-g1 Ra8-c8 Rg1-g4
e6-f5 (0:05:22,0:04:58)
f1-d3 (0:05:16,0:04:58) 0.14 9 402929 00:04 331247 (0) 9 0.14 Bf1-d3 Bf5-h3
Qd1-e2 Nd7-b6 Bd3-e4 h7-h6 f3-f4 h6xg5 f4xg5
f5xd3 (0:05:16,0:04:52)
d1xd3 (0:05:05,0:04:52) 0.16 9 426618 00:10 415830 (0) 9 0.16 Qd1xd3 Bf8-e7
Rh1-g1 Qd8-c7 Rg1-g4 Ra8-c8 Qd3xa6 Qc7xc2 Rg4xb4
f8-e7 (0:05:05,0:04:46)
h1-g1 (0:04:54,0:04:46) 0.03 8 620111 00:04 164562 (0) 8 0.03 Rh1-g1 O-O Rg
1-g4 Ra8-c8 Rg4-h4 h7-h6 Rh4xb4 Be7xg5
O-O (0:04:54,0:04:40)
O-O-O (0:04:37,0:04:40) 0.00 8 940194 00:03 351372 (0) 8 0.00 O-O-O a6-a5 Q
d3-b5 Qd8-c7 Qb5-c6 Qc7-d8 Qc6-b5
a6-a5 (0:04:37,0:04:35)
b3-d2 (0:04:35,0:04:35) 0.10 8 307874 00:03 134313 (0) 7 0.10 Nb3-d2 a5-a4
Nd2-c4 Ra8-c8 b2-b3 a4-a3 f3-f4
a5-a4 (0:04:35,0:04:18)
g1-g4 (0:04:35,0:04:18) 0.14 8 623534 00:03 149844 (0) 7 0.14 Rg1-g4 Nd7-c5
Qd3-c4 Ra8-c8 Kc1-b1 Rc8-b8 Nd2-e4
d8-a5 (0:04:35,0:04:01)
g4-h4 (0:04:27,0:04:01) 0.08 8 295002 00:04 156624 (0) 7 0.08 Rg4-h4 h7-h6
Rd1-g1 Nd7-c5 Qd3-c4 b4-b3 c2xb3 h6xg5 Be3xg5 Be7xg5 Rg1xg5 a4xb3 Nd2xb3 Qa5-e1
Kc1-c2 Nc5xb3 a2xb3
f7-f5 (0:04:27,0:03:51)
g5xf6 (0:04:19,0:03:51) 0.15 8 315123 00:06 250488 (0) 8 0.15 g5xf6 Nd7xf6
Nd2-c4 Qa5xd5 Nc4-b6 Qd5xd3 Rd1xd3 Nf6-d7 Rh4xb4 Nd7xb6 Be3xb6
d7xf6 (0:04:19,0:03:46)
d2-c4 (0:04:01,0:03:46) -0.07 8 842097 00:02 130264 (0) 8 -0.06 Nd2-c4 Qa5x
d5 Nc4-b6 Qd5xd3 Rd1xd3 Ra8-b8 Be3-f2 g7-g5 Rh4xb4
a5xd5 (0:04:01,0:03:30)
c4-b6 (0:04:01,0:03:30) 0.06 9 652556 00:12 577907 (0) 9 0.06 Nc4-b6 Qd5xd3
Rd1xd3 Nf6-d7 Rh4xb4 Nd7xb6 Be3xb6 Ra8-b8 c2-c3
d5xa2 (0:04:01,0:03:21)
b6xa8 (0:03:54,0:03:21) 0.48 7 260318 00:06 233382 (0) 7 0.48 Nb6xa8 a4-a3
b2xa3 Rf8xa8 Qd3-b3 Qa2xb3 c2xb3 b4xa3
f8xa8 (0:03:54,0:03:16)
d3-c4+ (0:03:48,0:03:16) 0.35 9 303100 00:06 303091 (0) 9 0.34 Qd3-c4 Qa2xc4
Rh4xc4 a4-a3 Kc1-b1 d6-d5 Rc4-c7 Be7-d6 Rc7-b7
a2xc4 (0:03:48,0:03:06)
h4xc4 (0:03:48,0:03:06) 0.29 11 526129 00:03 208109 (0) 10 0.29 Rh4xc4 a4-a
3 Kc1-b1 d6-d5 Rc4-c7 Be7-d6 Rc7-b7 Ra8-b8 Rb7xb8 Bd6xb8 b2-b3
a4-a3 (0:03:48,0:02:59)
c1-b1 (0:03:48,0:02:59) 0.41 10 360603 00:03 218709 (0) 10 0.41 Kc1-b1 d6-d
5 Rc4-c7 Kg8-f8 Be3-g5 Kf8-f7 Bg5xf6 g7xf6 Rd1xd5 Kf7-e6
d6-d5 (0:03:48,0:02:53)
c4-c7 (0:03:47,0:02:53) 0.55 10 339460 00:04 260752 (0) 9 0.55 Rc4-c7 Kg8-f
8 Be3-g5 Nf6-g8 Bg5xe7 Ng8xe7 Rd1-e1 Ne7-g6 Rc7-b7 Ra8-a4 b2xa3 Ra4xa3
g8-f8 (0:03:47,0:02:46)
e3-g5 (0:03:47,0:02:46) 0.53 8 282546 00:02 142892 (0) 8 0.52 Be3-g5 a3xb2
Kb1xb2 Ra8-a3 Bg5xf6 g7xf6 Rd1xd5 Ra3xf3
e5-e4 (0:03:47,0:02:39)
f3xe4 (0:03:41,0:02:39) 0.94 9 347566 00:04 221883 (0) 8 0.94 f3xe4 d5xe4 R
d1-d4 a3xb2 Kb1xb2 Ra8-a5 Rc7-c8 Kf8-f7 h2-h4
d5xe4 (0:03:41,0:02:30)
g5-f4 (0:03:35,0:02:30) 0.82 8 776138 00:03 642962 (0) 8 0.81 Bg5-f4 Nf6-e8
Rc7-b7 a3xb2 Kb1xb2 g7-g5 Bf4-e3 Ra8-a3
a8-a5 (0:03:35,0:02:16)
b2xa3 (0:03:29,0:02:16) 0.72 8 331294 00:01 91925 (0) 7 0.72 b2xa3 Ra5-f5 B
f4-d6 Be7xd6 Rd1xd6 Nf6-e8 Rc7-c8 b4xa3
b4xa3 (0:03:29,0:02:08)
c7-b7 (0:03:23,0:02:08) 0.68 7 288988 00:01 86550 (0) 7 0.67 Rc7-b7 Ra5-d5
Rd1xd5 Nf6xd5 Bf4-e5 e4-e3 Be5-g3
f6-d5 (0:03:23,0:01:58)
f4-d2 (0:03:13,0:01:58) -1.25 9 623486 00:05 312378 (0) 9 -1.25 Bf4-d2 a3-a
2 Kb1-b2 Be7-f6 c2-c3 Nd5xc3 Rb7-b8 Kf8-f7 Rb8-b7 Kf7-g6 Rd1-g1 Kg6-h5 Bd2xc3 Bf
6xc3 Kb2xc3 a2-a1Q Rg1xa1 Ra5xa1 Rb7xg7
a3-a2 (0:03:13,0:01:49)
b1-a1 (0:03:13,0:01:49) -2.28 12 493866 00:05 322471 (0) 11 -2.28 Kb1-a1 Be
7-f6 c2-c3 e4-e3 Rb7-b8 Kf8-f7 Rb8-b7 Kf7-e6 Bd2-e1 e3-e2 Rd1-c1 Ra5-a3 Be1-d2 B
f6xc3 Bd2xc3 Ra3xc3
e7-f6 (0:03:13,0:01:42)
c2-c3 (0:03:13,0:01:42) -2.46 12 417541 00:04 253301 (0) 11 -2.45 c2-c3 e4-
e3 Rb7-b8 Kf8-f7 Rb8-b7 Kf7-e6 Bd2-e1 e3-e2 Rd1-d4 Bf6xd4 c3xd4 Ra5-a3 Rb7xg7
e4-e3 (0:03:13,0:01:33)
b7-b8+ (0:03:13,0:01:33) -2.61 11 519754 00:03 181821 (0) 10 -2.61 Rb7-b8 Kf
8-f7 Rb8-b7 Kf7-g6 Rd1-g1 Kg6-h5 Bd2-c1 e3-e2 Bc1-d2 Nd5xc3 Rb7-b2 Ra5-f5 Bd2-c1
f8-f7 (0:03:13,0:01:25)
b8-b7+ (0:03:13,0:01:25) -3.56 11 478416 00:00 24420 (0) 10 -3.55 Rb8-b7 Kf7
-g6 Rd1-g1 Kg6-h5 Bd2-c1 e3-e2 Rb7-b3 e2-e1Q Rg1xe1 Bf6xc3 Rb3xc3 Nd5xc3
f7-g6 (0:03:13,0:01:18)
d2-c1 (0:03:13,0:01:18) -3.50 11 363774 00:02 152271 (0) 10 -3.49 Bd2-c1 e3
-e2 Rd1-g1 Kg6-h5 Rb7-b3 Bf6xc3 Rb3xc3 Nd5xc3 Bc1-d2 Ra5-c5 Rg1xg7
e3-e2 (0:03:13,0:01:11)
d1-g1+ (0:03:13,0:01:11) -6.62 10 377171 00:05 331271 (0) 10 -6.61 Rd1-g1 Kg
6-h5 Rb7-b2 Nd5xc3 Rg1-e1 Nc3-d1 Re1xe2 Ra5-c5 Bc1-f4 Nd1xb2 Re2xb2
g6-h5 (0:03:13,0:01:05)
b7-b2 (0:03:13,0:01:05) -6.62 10 291601 00:01 60594 (0) 9 -6.61 Rb7-b2 Nd5x
c3 Rg1-e1 Nc3-d1 Re1xe2 Ra5-c5 Bc1-f4 Nd1xb2 Re2xb2
d5xc3 (0:03:13,0:00:55)
c1-d2 (0:03:13,0:00:55) -8.01 10 562244 00:03 183944 (0) 9 -8.00 Bc1-d2 Ra5
-d5 Rg1-e1 Nc3-b1 Bd2-c1 Rd5-d1 Re1xe2 Rd1xc1 Ka1xa2 Nb1-c3 Ka2-a3 Nc3xe2 Rb2xe2
a5-a4 (0:03:13,0:00:47)
b2-b6 (0:03:07,0:00:47) -12.36 8 315015 00:01 89297 (0) 7 -12.36 Rb2-b6 Bf6
-d4 Rb6-a6
f6-d4 (0:03:07,0:00:40)
b6-b4 (0:03:03,0:00:40) -18.32 9 560106 00:01 451206 (0) 9 -18.32 Rb6-b4 Ra
4xb4 Bd2xc3 Rb4-b1 Ka1xa2 Rb1xg1 Bc3xd4 Rg1-g2 Ka2-a3 e2-e1Q
a4xb4 (0:03:03,0:00:32)
d2xc3 (0:03:03,0:00:32) -39.29 11 481166 00:03 226087 (0) 10 -39.28 Bd2xc3
e2-e1Q Rg1xe1 Bd4xc3 Ka1xa2 Bc3xe1 Ka2-a3 Kh5-g4 Ka3-a2 Kg4-h3 Ka2-a3 Kh3xh2
d4xc3 (0:03:03,0:00:24)
a1xa2 (0:03:03,0:00:24) -41.30 12 0 white must play Ka1xa2
b4-a4 (0:03:03,0:00:19)
a2-b3 (0:02:58,0:00:19) -19.75 9 280428 00:00 70625 (0) 8 -19.75 Ka2-b3 Ra4
-a7
a4-a1 (0:02:58,0:00:14)
g1-g3 (0:02:53,0:00:14) -30.82 9 229830 00:02 111119 (0) 8 -30.81 Rg1-g3 e2
-e1Q Rg3-h3 Qe1-h4 Rh3xh4 Kh5xh4 Kb3xc3 Ra1-a2 Kc3-d4 Ra2xh2
e2-e1Q (0:02:53,0:00:06)
g3-h3+ (0:02:53,0:00:06) -34.55 8 373524 00:05 321869 (0) 8 -34.54 Rg3-h3 Kh
5-g4 Rh3-g3 Qe1xg3 h2xg3 Bc3-b4 Kb3-c4 Ra1-d1 Kc4-b5 Kg4xg3
h5-g4 (0:02:53,0:00:00)
h3-g3+ (0:02:53,0:00:00) -34.55 8 486681 00:00 405 (0) 7 -34.54 Rh3-g3 Qe1xg
3 h2xg3 Bc3-b4 Kb3-c4 Ra1-d1 Kc4-b5 Kg4xg3
g4-f5 (0:02:53,0:00:00)
g3-f3+ (0:02:48,0:00:00) -34.23 7 245311 00:03 175186 (0) 6 -34.22 Rg3-f3 Kf
5-e4 Kb3-c4 Ke4xf3 h2-h3 Qe1-e6 Kc4-d3 Ra1-c1
f5-e6 (0:02:48,0:00:00)
f3xc3 (0:02:43,0:00:00) -499.99 8 337889 00:03 213232 (0) 7 -MATE05 Rf3xc3
Qe1-d1 Rc3-c2 Qd1-d3 Rc2-c3 Qd3-b5 Kb3-c2 Ra1-a2 Kc2-c1 Qb5-f1
e1-d1 (0:02:43,0:00:00)
b3-b4 (0:02:43,0:00:00) -499.99 10 474659 00:05 425939 (0) 9 -MATE04 Kb3-b4
Qd1-d4 Kb4-b5 Qd4xc3 h2-h4 Ra1-b1 Kb5-a4 Qc3-b4
d1-d4 (0:02:43,0:00:00)
b4-b5 (0:02:43,0:00:00) -499.99 17 254655 00:05 237017 (0) 16 -MATE03 Kb4-b
5 Qd4xc3 h2-h4 Ra1-b1 Kb5-a4 Qc3-b4
a1-b1 (0:02:43,0:00:00)
c3-b3 (0:02:39,0:00:00) -499.99 66 156229 00:04 154892 (0) 66 -MATE02 Rc3-b
3 Rb1xb3 Kb5-a6 Qd4-b6
b1xb3 (0:02:39,0:00:00)
b5-a6 (0:02:39,0:00:00) -500.00 100 104 00:00 104 (0) 100 -MATE01 Kb5-a6 Qd
4-b6
At thursday i think it was when Weiner became real nervous. Also
Friedel arrived at end of the week. During a game i very suddenly
saw him for the first time. After i said 'hi' to him and hurried
myself back to my board, meanwhile Friedel was LOUDLY demanding
that i had to shake his hand. This was however not possible for me.
When I see Friedel in a reflex my right hand goes down to feel whether
my wallet is still there.
Weiner was that nervous that he also took a look at the other boards.
At the 6th round DIEP played against Tiger. Now people who read this
must realize that the day before this there an exhausting blitz
event, and that when i started DIEP for the tiger game, that i didn't
very carefully check what i had done the day before that. Diep played
with a book i had prepared for white against Nimzo/Fritz to face Najdorf.
Now i had black and even though i had spent a lot of time the days before
to prepare Gruenfeld with black, all this time was wasted now, because
i forgot to do 2 things:
- turn on the egtb
- turn on the right tournament book.
This might seem real stupid to everyone i bet, but consider the time at
which i got up this day: 7 AM, and that end of august.
Also i'm in textmode, i can't see what book gets selected. Having a
good interface showing those things in a better way, or simply having
a tournament which is made BEFORE the world champs start, that's in general
much smarter.
I'm sure that the game against tiger would have gone different when playing
with the prepared gruenfeld lines. Now DIEP played only with the huge book,
which selected Kings indian defense. This very clearly showed me again
that one should play all openings in testgames. At the ICC diep plays quite
some openings, but i can't remember i ever studied a K.I.D. game played by
DIEP at the black side.
Now diep had a position where it
a) didn't care for the queen side
b) didn't care for the king side
It just cared about minor details. Especially the lines of both programs
was very disgusting. Neither of them had plans to play ever g5. Both of them
planned to play h5-h4 and block with white with h3. Also both showed Qg5 in the
mainline. I was really amazed that Tiger searched this deep here. It searched
each move between 15 and 17 ply, and that sure wasn't because it prunes that
much.
DIEP also searched deep. It proofs again how well knowledge works. If you don't
know about the king side setup for KID, then you can shake it. Despite the
search difference, both programs showed similar lines in the first phase of the
game, incredible stupid lines actually. Basically they were both caring
only for the queen side instead of white attacking queen side and black
attacking
king side. Only the shot R1b6 was missed by DIEP. At that time the game was
already
decided a long time ago. Basically the game was decided already by book. Diep
got
out of book with -0.80, so even with the right plan it's questionable what would
have been the result of the game, as i don't see tiger lose because of a
tactical
trick when getting 15 to 17 ply.
I see K.I.D. as a big swindling attempt from black, which objectively can't be
good at all. White gets so many tempi. Having the right plan with white:
collecting material and taking action queen side, i'm sure that it's hard
to beat a Tiger searching 15-17 ply in this position for *any* human.
A program with the right plan in my eyes is simply very hard to beat. Especially
if you don't use the vague 'opponent modelling' approach. This is basically
the problem with GMs. They're so much used to take into account that they
must play objective chess, that when they play against an opponent which can
be fooled as it's not clever, that they don't know how to act as they don't
know HOW a program basically is searching and WHAT it is using to evaluate
lines.
Using these definitions and taking into account that DIEP also was bad
after book, DIEP lost this game chanceless. If a program
- do not have the right plan
- see the position in the same way as your opponent
- misevaluate the position
- gets outsearched by 2 to 3 ply
- gets bad out of book
Then you lose with natural induction in my opinion.
Again i continued copying the EGTB files. With a 56k6 modem paying a few dollars
an hour for internet, it's not easy to download all 5 men. Even copying them
from
a CDrom with a 52 speed CD player at my Maxtor 30Gb harddisk it's not copied
quickly. Actually i started copying already at Monday morning. Matthias had then
borrowed me his CD set and i started copying basically the 3 against 2
databases.
Now on thursday i borrowed from Stefan 3 against 2 CD set, as i still hadn't
finished copying it! All together it took many hours to copy all of them. In
total the EGTB take 5.2 gigabytes (i didn't divide this
by 1024^3 but by 1000^3).
That's huge! Note that at Matthias CDroms there were even more EGTBs: 4 against
1,
for those who can ftp stuff: take a look at the 6 men directory at Bob's
machine.
Just starting DIEP with the 5 men takes quite a bit of time and a huge amount of
RAM
ONLY for indices already. About 15mb in DIEP's case (3, 4 and 5 men but not
using 4 against 1 egtbs).
Compressed Nalimov really isn't user friendly yet as to this amount one has to
add the endgametablebase cache. By default i set this at 1 mb, as everything
gets into hashtable anyway during search. Putting it at bigger values are
when talking about DIEP really a waste of time i measured.
Note that at the time of writing 5.2 Gb is also pretty hard to get. The fact
that
i didn't have a single 5 men except KRPKR at my harddisk already says enough
about this.
At this time in the tournament i realized that the tournament books from Kure
weren't that a big problem to avoid in the future. Basically if you play what is
inside NCO 1999 (nunn's chess openings) then you already don't get killed by
Kure. With a few exceptions as some lines are called 'unclear' where they
inreality
sometimes are a forced win for either side.
Basically you get completely killed if you play an outdated line, usual not in
the
book as being very good.
Like the Nimzo-Shredder game. This is inside NCO as being a huge advantage for
white.
The real surprising book is always Noomen. His preparation is more into detail.
Some lines from the Kure book get killed completely by Noomen. Like Fritz-Tiger
shows a real killerline.
However amazingly Tiger can't win it.
Basically there are just 2 good books in computerchess. Noomen and Kure book.
The
shredder book is already having a lot of outdated lines. Noomen has the bad luck
that he doesn't have programs which can play according to the lines.
Also some 'refutations' which give humans great chances to win are not exactly
refutations against a computer.
This is of course the good work of all these gentlemen, they are real good in
preventing positions where a program plays completely bad.
So if i say: "NCO solves the problem" then take this in the right way. It takes
years to figure out which lines a program can play and what it can't. Also the
huge number of so called 'unclear' lines in NCO surprises me. Many are simply
dead won for either white or black.
The huge difference with the nowadays theory and the old lines is amazing. Many
lines are completely refuted. Automatically generated books have 0% chance
against
this.
The WMCC also shows how much engines still have to improve. If you're won out of
book like Nimzo against Shredder, how in the world can you not win it then?
I mean let's not complain too loud, but a PIECE up?
How did Shredder survive this? Preventing Qxb4 is still leading
into a position with a piece down, let's face it.
Predicting the outcome of games is gonna be real easy like this:
- nimzo wins with passed pawns
- nimzo does not win against passed pawns as it's too afraid for them
When i mentionned this 'opponent modelling' prediction during the game
Stefan laughed at me! How happy he was a few hours later!
Even worse as shredder was Xinix passed pawn evaluation against DIEP.
The bad thing is that DIEP went later out for a pawn at h7. This was real
stupid. Rh5-h6xh7 is real bad. That rook is out of play simply. Now
Xinix can draw. Rd3! is real good. Even dual junior takes 1.5 minutes here
to see Rd3. Amazing is that after Rh8-c8-c1 that black has a draw with a
pawn less. Too bad for Xinix that it missed it against DIEP.
It's amazing that all programs i tested so far fall for the drawing joke.
The drawing joke is basically giving away the h7 pawn. This with a piece
up for DIEP. No need to capture that pawn.
After d5? from xinix it was of course game over. Diep expected e1Q here
with a near to draw score. After d5 i had directly +3 getting up
to +10 real soon.
The interesting thing is that just focussing attention to a single area,
like overestimating passers too much, that this can be rewarding.
Again a real interesting game was Diep-Crafty. Always a quick loss or a quick
win. From now on i'll only bet on the last. Where DIEP's book was not NCO
improved yet, crafties WMCC book was simply not existing.
To my big amazement crafty played Najdorf. Now this is a big surprise for
two reasons.
- crafty sucks tactical in najdorf as it doesn't do checks in qsearch
- crafty doesn't have a book to prevent direct losing against a mainline
The amazing thing was that after poisoned pawn got onto the board, that crafty
was so quick out of book. Even an automatically generated book still has the h5
move in book. Note that this move is not too difficult to find if you do
checks in the qsearch with a reasonable king safety. Diep needs 12 ply to find
h5. Just because of slowness of DIEP this takes 2.2 minutes to fail low on O-O
and in total 9:51 to get to the h5 move.
In future versions that will be hopefully sooner. Still after castling DIEP
also was out of book. Just showing the scores then says enough:
e2-e4 (2:00:00,2:00:00) book
c7-c5 (2:00:00,1:59:19)
g1-f3 (2:00:00,1:59:19) book
d7-d6 (2:00:00,1:59:06)
d2-d4 (2:00:00,1:59:06) book
c5xd4 (2:00:00,1:58:48)
f3xd4 (1:59:59,1:58:48) book
g8-f6 (1:59:59,1:58:21)
b1-c3 (1:59:59,1:58:21) book
a7-a6 (1:59:59,1:58:02)
c1-g5 (1:59:59,1:58:02) book
e7-e6 (1:59:59,1:57:44)
f2-f4 (1:59:59,1:57:44) book
d8-b6 (1:59:59,1:57:29)
d1-d2 (1:59:59,1:57:29) book
b6xb2 (1:59:59,1:57:16)
a1-b1 (1:59:59,1:57:16) book
b2-a3 (1:59:59,1:56:53)
f4-f5 (1:59:59,1:56:53) book
b8-c6 (1:59:59,1:56:28)
f5xe6 (1:59:59,1:56:28) book
f7xe6 (1:59:59,1:55:59)
d4xc6 (1:59:59,1:55:59) book
b7xc6 (1:59:59,1:55:43)
e4-e5 (1:59:59,1:55:43) book
d6xe5 (1:59:59,1:51:36)
g5xf6 (1:59:59,1:51:36) book
g7xf6 (1:59:59,1:50:43)
c3-e4 (1:59:59,1:50:43) book
f8-e7 (1:59:59,1:46:38)
f1-e2 (1:59:59,1:46:38) book
O-O (1:59:59,1:43:49)
b1-b3 (1:55:39,1:43:49) -1.40 12 10410289 02:08 5131791 (0) 11 -1.40 Rb1-b3
Qa3xa2 Rb3-g3 Kg8-h8 O-O Rf8-f7 Qd2-h6 Bc8-b7 Be2-h5 Be7-f8 Ne4-d6 Rf7-c7 Qh6xf
6 Bf8-g7
a3xa2 (1:55:39,1:40:02)
O-O (1:55:12,1:40:02) -0.95 12 10069052 03:13 7721131 (0) 12 -0.94 O-O Kg
8-h8 Qd2-h6 Rf8-f7 Rb3-g3 Bc8-d7 Be2-h5 Ra8-f8 Bh5xf7 Rf8xf7
g8-h8 (1:55:12,1:35:22)
d2-h6 (1:55:12,1:35:22) -0.02 12 11084226 04:17 10175948 (0) 12 -0.02 Qd2-h
6 Rf8-f7 Rb3-g3 Ra8-a7 Be2-h5 Qa2-c4 Qh6-e3 Be7-f8 Bh5xf7 Ra7xf7 Ne4-g5 Rf7-b7 R
f1xf6
f8-f7 (1:55:12,1:31:05)
b3-g3 (1:55:10,1:31:05) 0.08 12 10434337 00:44 1857642 (0) 11 0.08 Rb3-g3 R
a8-a7 Be2-h5 Qa2-c4 Bh5xf7 Be7-c5 Ne4xc5 Qc4xc5 Qh6-e3 Qc5xe3 Rg3xe3 Ra7xf7 Re3x
e5
a8-a7 (1:55:10,1:30:13)
g1-h1 (1:51:41,1:30:13) 1.89 12 10632112 00:30 3252994 (0) 11 1.88 Kg1-h1 Q
a2xc2 Be2-h5 Be7-f8 Bh5xf7 Ra7xf7 Qh6-h5 Rf7-e7 Ne4-d6 Bf8-g7 Nd6xc8
a2xc2 (1:51:41,1:26:00)
h6-h5 (1:51:36,1:26:00) 4.56 12 10335746 03:18 8038735 (0) 12 4.55 Qh6-h5 R
f7-f8 Be2-d3 Qc2xd3 Rg3xd3 f6-f5 Rd3-h3 Be7-d8 Ne4-d6 Bd8-f6 Qh5-h6 Ra7-f7 Nd6xf
7 Rf8xf7
e7-f8 (1:51:36,1:20:27)
e2-d3 (1:47:16,1:20:27) 9.73 12 10345462 03:11 7465878 (0) 11 9.73 Be2-d3 Q
c2-c3 Ne4xc3 f6-f5 Nc3-e4 a6-a5 Ne4-g5 Rf7-e7 Ng5xh7 Re7xh7 Qh5-g6
c2-c3 (1:47:16,1:17:54)
e4xc3 (1:45:35,1:17:54) 10.71 13 11309369 01:03 3009384 (0) 12 10.71 Ne4xc3
f6-f5 Nc3-e4 a6-a5 Ne4-g5 Rf7-e7 Ng5xh7 Re7xh7 Qh5-e8 Ra7-f7 Qe8xc8 e5-e4
f6-f5 (1:45:35,1:08:48)
f1-d1 (1:45:35,1:08:48) 14.34 14 25902144 08:35 24433268 (0) 14 14.34 Rf1-d
1 Rf7-f6 Qh5-g5 Rf6-g6 Qg5-d8 Rg6xg3 h2xg3 Ra7-f7 Qd8xc8 Rf7-f6 Bd3xa6 e5-e4 Rd1
-d6 Rf6-f7
Graham tested crafty at the O-O? position but crafty no way ever plays h5
itself.
This is real interesting and again showing that the importance of book for
chessprograms is basically there because programs do so bad themselve in
opening.
The pairing of the 9th round took quite some time. This had a reason. First of
all the previous rounds the pairing was messed up. Before the 8th round
v/d Herik even had voted in a very uncommon way. Instead of giving everyone
a vote he gave only a few people a vote. I was for example denied voting
just like crafty and some other programs from who he feared opposition, despite
that all these programs could still win the amateur world title. V/d Herik
allowed only a few to vote. Weirdly Both Frans Morsch was on the right of v/d
Herik now and then raising a hand for Fritz and on the left of v/d Herik
Matthias Feist was now and then voting for suggestions as well. Matthias Feist
operated Fritz. Lucky Fritz didn't have a 3d vote
by means of Friedel in the middle, as it was 9.30 AM. Friedel would
arrive in the room hours later. Weiner was there to vote for Shredder.
Stefan seems not allowed to vote on his own about shredder. Stefan was
elsewhere in the playing hall.
The voting and suggestions from v/d Herik as well as his pairing was a big mess
actually. In the end he did what he wanted, now and then counting votes from
chessbase double and not counting my nor Grahams vote: "as you guys are not
allowed to vote". Next time just tell us to get there at 10 AM Jaap
instead of 9 AM, but never again call this democratic.
This was a waste of my time and energy, and it sure wasn't democratic.
After playing many computer chess tournaments with many real bad pairings
i was even amazed by the next pairing a couple of hours later for the 9th round.
I protested bigtime when v/d Herik was again toying with paper cards with names
of the programs at it.
Crafty was paired with white against Fritz. Now i already had told v/d Herik
my opinion that this wasn't a good idea. Also the fact that Sos played
against Xinix and that Zchess faced Fritz was not a good idea.
I volunteered to play against Fritz. Note that i had prepared something
against Fritz, i wouldn't repeat the game as against Shredder for sure.
Against 1.d4 i had prepared real good now the Gruenfeld by writing over NCO
lines bigtime.
To start with the most important: SOS had 4.5 points, so had Zchess. Both
didn't play each other. The winner would be amateur world champion. Now
it was already clear that Zchess was a real good program. Fritz didn't
play it yet, but shredder was forced to play it now.
Shredder had played DIEP. DIEP hadn't played Fritz.
People must take into account that for the 8th round JUNIOR was allowed to not
play Pacque expert because that would be bad for its SOP (sum of opponents
points,
in german called Buchholz, in dutch weerstandspunten). So if junior would
win the lottery, and others would do something incredible stupid then junior
would still have a small chance to finish equal with other programs as divided
first after which it would make a bigger chance to be with the first 2 who would
play a tie.
but now by this decision for the 9th round the WORLD TITLE directly was given
to SOS. Note that i reallylike rudolf huber, but this is weird decision taking.
Someone is afraid for his SOP, completely unimportant compared to the
number of points you have but the complete pairing gets changed because of
that, but here zchess was DENIED to play against SOS
direclty for the world title. So the amateur title has been completely given
away by this decision, even though there was a real good solution possible.
The only problem would be that DIEP would face Fritz and that chessbase very
possible
would not get back home with a single world title.
Now diep had an equally number of points as crafty. So pairing DIEP against
Fritz
would be no problem.
In tournaments DIEP has a reasonable record against Fritz and Shredder. You can
easily look that up. Only last tournament DIEP lost in the dutch open. You can
see that this was a B99 game where DIEP lost for a similar reason as why crafty
lost twice from Fritz&Nimzo in the WMCC. I had learned my lesson there.
I had prepared against Fritz for this tournament.
Crafty however had lost chanceless on book against Nimzo with white.
It would play against the EXACTLY same book against Fritz now for sure,
a sure zero again. 100% chance. Now i
WARNED Graham that it was NOT a good idea to again play the same line
against Fritz.
"learning will prevent that, and don't tell them i have a different book too".
But he did not CHECK whether in that different book a different line
would get played, despite warnings that it would happen. If you use an
automatically
generated book there is 100% chance you lose against the Kure book in the
Najdorf,
you can only hope the engine crashes somewhere playing thereby a help mate in 1
move.
In fact this 'new' crafty book was even more pathetic as against Nimzo.
It didn't even know the Rg1-g7 moves.
I don't need to mention that it lost in the 9th round in
exactly the same mainline against
Fritz as it did against Nimzo.
Now suppose Graham had not ignored my warning, and he would have been more
fanatic
to NOT lose the same line for the second time. I'm SURE it would not have
mattered,
because how in the world could he change the crafty book within say 5 minutes?
He would NEVER have managed this.
Now one can make mistakes. One can get killed easily by the kure book,
but losing 2 games for the same reason?
- the game against rebel was exactly played before
- the game against Fritz was exactly played against Nimzo, now
crafty even sooner out of book.
- the game against DIEP was lost because of the same Najdorf problems
- etcetera
Perhaps i may suggest something for crafty: NEVER EVER AGAIN PLAY NAJDORF IN
A WORLD CHAMPIONSHIP, with white NOR with black.
Note that it played Najdorf old mainline against Nimzo/Fritz, not
a very good idea if you're using nullmove without checks in qsearch,
as that means you tactical search 9 ply versus Fritz/Nimzo 4 ply more.
Fritz could even affort to capture a pawn at h2 and still win chanceless
with induction to the search and position.
So even IF the book would be at equal level, it still would get outsearched
pathetically.
CONCLUSIONS/THOUGHTS/ORGANISATIONAL THINGS
I find Shredder really deserved won the title. It had lost positions after book
against
Fritz and Nimzo but still drew them. It basically won the title because
it has the best far endgame of all programs, no program gets even close to that
right now.
One of my plans is obvious: improve the far endgame in DIEP
the whole book issue is also clearly defined. Programs are not blundering
too quick, but books are simply getting into real bigtime lost lines.
A human who doesn't blunder too quick is NOT getting into those real bigtime
lost lines as he has a few of new books and the lines he plays on average
aren't lost bigtime.
Regrettably I saw only 2 books this tournament that could be called 'reasonable'
to human standards. 2200-2400 standards that is, we're not even talking about
2500+
standards. Namely the book by Alexander Kure and the Jeroen Noomen book.
The so called amateur books and also the shredder book got killed just
too easily by new theory, shredder and the insomniac book a bit less as the
rest but still...
Just writing over the mainlines from a book like NCO
would already have prevented big disasters for especially crafty, diep and
shredder would have scored a few points more against the Kure book.
Especially the last game against Francesca i could mourn a lot about not doing
this, as diep played the horrible Ng4? move, instead of the NCO Nd5 move. Note
that i had a fail high to Nd5. At a dual DIEP would already have played it.
However if i instead of doing nothing had just written over a few lines against
the popi jopi (popular outdated) books then DIEP would sure have done better
this game. That DIEP saved the endgame to a draw shows that the combination of
just a bit of endgame code and EGTB already mean a lot.
I'm not unhappy about DIEP's performance. Basically a lot of broken passed
pawn code in DIEP performed real bad and DIEP just like crafty it needs a
smarter operator or a better book&interface.
DIEP operator problems:
- i forfeited in a better position against junior
- i forgot to select the tournament book against tiger
- i needed to copy all EGTB during the tournament instead of
before the tournament.
- 4 weeks before the WMCC i not only started to rewrite all kind of
things in DIEP's evaluation, but also have been busy with
the win32 interface a lot. This has resulted in a real
cool datastructure and a lot of new code for the WMCC, but into
disaster against
Zchess: c4?? was only caused by the broken passed pawn code
in DIEP, and the win32 interface wasn't ready yet for wmcc.
Basically the old rule: only play with a well tested
version is what i should have followed.
What i should have done: write over lines out of NCO the last 4 weeks
before WMCC and not care for the engine nor interface.
A good operator is really needed to get a reasonable result in a world champ,
though a good interface is usually solving this problem, because note that
both crafty and diep ran this WMCC in textmode.
Now Graham has now some experience for next WMCC, i'll have a graphical
interface
hopefully working correctly at the dutch champs, perhaps that's already
interesting.
The level of games at a WMCC is still not very high. This is partly because
the books are pathetic, and the evaluations of programs have huge gaps.
It's real interesting that Shredders superb far endgame play (when compared
to the quality of others) is enough to win the title.
Even getting lost out of book is no problem then, but winning is something
different.
For the far future it's amazing to see how simple it is to draw commercial
engines. This is already the case in computerdraughts. The big lack of
knowledge in programs there gives every engine a chance to draw easily.
Much easier as humans can draw each other. Will this in computerchess
be the case too?
The good thing to realize is that having a good engine is still more important
as having a good book. Also the definition of a good book seems to be much
more simple as i supposed. Just write over a new book that's it.
So assuming many will follow this principle, as i'm most likely not the
only one who figured out how big the gap between the new books and the
automatic books was, in future of computerchess the best engine will simply
win.
Right now Shredder is superb into exchanging into a draw from a lost position
and also great into trying to win endgames (like it managed to win this
endgame against DIEP), so Shredder clearly won the title deserved. The half
point difference doesn't even reflect the real difference. If we eliminate
the book factor, then the difference would be much larger.
However as long as programs like crafty can't find the h5 moves itself,
then book will still dominate against them.
If i would see this position for the first time, then there i would find h5
myself. I'm not even IM yet. Right now i'm rated 2254 at the FIDE list. Even
though this will real soon get up to 2300+ up to 2400 perhaps, this is
still way below the average level a program CAN show.
This means simply that the big knowledge lack of programs is really showing
nowadays. Just do a silly move in such a line first and programs will not
invent the plan itself.
In this respect i am very happy for DIEP's future, as it already managed
to save a lot of games because of a bit of extra knowledge. Like it
managed to save a lost middlegame against Shredder to a drawn endgame;
the far endgame where diep's eval is still pathetic it lost (so logical).
Diep saved a dead lost position against Insomniac, it especially directly
evaluated its chances quite accurate. Also a big save happened against
Francesca. From a dead lost middlegame/endgame it managed to get into
a complicated rook endgame with just a single pawn down,
where draw was way easier to achieve. Against crafty DIEP won of course
easily (it was real easy to mate crafty in this positoin), but when analyzing
after the game it appeared that DIEP would have played the poisoned pawn
correctly with black, this basically based upon a reasonable king safety
evaluation.
The only dissappointments was that hardly anything had happened at book
and especially the fact that DIEP didn't manage to win the middlegame
position against junior. I blame the broken passed pawn evaluation for that,
but this is a bit dubious.
Also dissappointing was the a5-a3? manoeuvre against Francesca, the h5?! move
against Insomniac and the c4?? move against Zchess.
A good example of a tough plan is also the sveshnikov line diep played.
If you don't have Ra8-a7-f7 plan as engine you can shake it, where for a human
this is very easily to figure out.
Also interesting is the evaluation aspect of the Re4-Rxh7? idea in DIEP-Xinix.
Even though i had more things to do as just to watch the game, i was as a human
already scared like hell when i saw the Rh4 move. "DIEP is a piece up, why
not win it easily instead of grabbing another pawn?" Such simple statements
are real hard to put into an evaluation however.
As soon as those problems get solved however domination of computers in chess
will be clear. Right now i can fool each program too easily too.
Basically any chess player who doesn't give away a piece each few moves can
still do well against a chessprogram. The human-computer challenge is
far from clear as long as these big eval bugs are there.
On the other hand it's clear that a program can beat any human if he doesn't
take into account what kind of child he's playing. Also he loses if he still
thinks he's playing a baby.
From this viewpoint still a lot must be achieved in Artificial Intelligence!
Doesn't take away the question who is gonna organize the computerchess a bit
better next years ICCA tournament, as Levy will be most likely even more busy,
but he sure wants to have the computerchess tournament over there as with
entry fees of 100,250 and 500 dollar and hiring of monitors for 1/3 of their
price it sure is an interesting event to keep, especially when programs
where at most 10 lines of assembly have changed from since 1990 need to pay
250 dollar in order to lose 9 games in a row.
If you are organizing a big event like the MSO games, you can't make it clear
to me that you didn't already hire this hall, where the computerchess
part of it only took about 1/10 of the hall space (and no extra space for
the few visitors. Extra personnel wasn't needed as
v/d Herik and Don Beal solved the problems. The hire for the monitors was done
using a company who tried to earn something, so no problems there either.
Considering that even for small games you had to pay 13 ENGLISH POUNDS and
up to 25 pounds entry fee for all kind of other small games, which i use
to play for free here at home.
Note that i found it real bad that apart from Andrew Williams, Richard Lang and
Jonathan Schaeffer i hardly counted visitors for the WMCC, unless you see
Friedel and Weiner as visitors...
Perhaps CCC is not making enough commercial for computerchess?
Perhaps RGCC was a better form of propaganda for computerchess?
Sure is that at the WCC in paderborn there were a hell of a lot more
visitors. Note that the analysing chessbase GM perhaps also made quite a
lot of difference.
Sure is that with fees of 100,250,500 the icca could easily affort to hire
one of the 6 GMs or dozen IMs that attended the sponsored chess tournaments over
there
to give comments each day on the computerchess games. They were there
anyway.
I'm sure that most participants even missed that just 30 metres away from
them in the hall 6 GMs and about 10 IMs were playing rapid chess over there
each afternoon. Amazing in this was that the computerchessevent with 14
participants
was the first few rounds paired using a program, where these 50 chessplayers
were
paired by hand!
Note that probably only 50 chessplayers joined the rapid chess as the GMs/IMs
were sponsored by a company to get there, but the normal chessplayers had to pay
entry fees of 25 pounds JUST to play a few rapid chess games and the rules
for the tournament said: "regrettably we don't have prize money yet, but if
there are more as 60 players then there will be a good prize money". 50 players
joined. Lowest rated player i saw was a single person rated 1700,
then within a few players very quickly getting up to over 2000. Where were
all those 1000 to 1700 rated players i usual see joining tournaments en masse
when so many titled players join in a sponsored chess tournament?
Note that from all those chessplayers hardly anyone walked 30 metres further
to watch the computers, as in contradiction to all other tournaments the
computerchess didn't have its own board saying: "world championship
computerchess".
So if you would take a picture of the hall, you would see a few computers, but
sure miss that it was a world championship. So from all the hundreds of people
that were in the alexandra palace to join the games, no one COULD KNOW that
there
was a world micro computerchess championship over there.
Also when i get to the homepage from MSO i have huge problems finding the
computerchess area over there.
Vincent Diepeveen
This page took 0.05 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.