|
1.
Please visit our "little" News-Ticker in
User Files, Engines
selection.
Here can be found information about "NEW" User files / updates too!
Go to
News-Ticker:
Overview ,
Main Page ,
Page 1,
Page 2,
Page 3,
Page 4,
Page 5,
Page 6,
Page 7,
Page 8,
Page 9
April 04th, 2005
New version of the opening book by Gabriel LEPERLIER
95,
Gabriel LEPERLIER
Gabriel sent today
a new version of his opening book. Each month on one time he do that and more
and more Gladiators seems to like the opening book by Gabriel a lot. I can see
it on our webpage statistics. You can find the opening books in our "User Files,
engines" selection!
THANKS Gabriel
Text by Gabriel LEPERLIER
"I'll try now to
make updates on a regular basis : 1 per month. I only add my engine vs engine
games published on "Le Fou Numerique" to this book. My tournaments are usually
made on Arena's GUI, but I am using ChessBase GUI too. My Arena's favorite
option is the one that allow to take a random opening in a PGN at move X. As I
use GMI and IM PGN's, the barebone of my book is solid. But I am playing my
tournaments with all Free UCI engines, the weaker same as the best ones, so we
can say that those engines gives diversity to my book. Making a tournament with
my book gives a real varienty of play. Using my book as a tournament book to
take advantage vs another book is a interesting idea because you have many
chances to take your oponent out of book quickly ;-)."
April 04th, 2005
Learn more about Pharaon
Interview with Franck ZIBI (Pharaon)
94,
Christopher Conkie, Michael Diosi, FQ


v. 3.2, 2.62
WB / UCI compatible, chess tutor
Home of
Pharaon
Hint: Pharaon is number 27 in ATL-1 Rating-List!
In the latest week Christopher Conkie,
Michael Diosi and myself are working on interview questions for
Pharaon programmer Franck ZIBI. Pharaon 3.2 is around three months
available and it's very interesting to know more about the development by Franck
because newer versions of Pharaon are a longer time not available!

The very sympathic French
programmer Franck ZIBI.
Here in one of the latest in France organiced tournament.
***
Note: Pharaon started for four days in ATL-2 ***
ATL League
I learn a lot about chess from different mails
Franck have written me in the lastest six years. The interview with him is very
interesting and can be found on the webpage of Christopher Conkie and
Michael Diosi. Here can be found the Arena / Pharaon setup too. This
interviews is a cooperation with Franck ZIBI / Exacta Chess and Arena Chess GUI.
Exacta Chess
Christopher Conkie wrote me:
Hello Everyone,
It’s me again….your old friend Chris….. ;-)
We have been a little busy of late.
We have a new Interview section at the Exactachess website.
Our first interview is with Frank Zibi, the author of Pharaon.
Go now……..read…….here………
http://www.exactachess.com/modules.php?name=Interviews
Very interesting.
Regards, Christopher
I will give short comments to
two of my interview questions with Franck ZIBI.
Comments
to the answer of questions 10 by Franck:
Chess 960 is one of the most interesting things in computer chess today. In
my interviews I asked the programmers for Chess960 support. You can find in all
the interviews I created comments by other programmers to Chess960. Arena Chess
GUI is since two years compatible to Chess960 (FRC = Fischer Random Chess) and
Shuffle chess. The other GUI programmers blocked the Chess960 support or have
for some reasons no interest to add Chess960.
ChessTigers:
http://www.chesstigers.de
(Chess960 WM organization with a fantastic news page)
Smirf:
http://www.chessbox.de/Compu/fullchess1.html (information to Chess960)
Comments
to the answer of questions 13 by Franck:
I created an interview with the programmer of Fruit for around 2 weeks. In
this interview you can find around 15-16 questions with some comments from the
main tester of Fruit (Rober Allgeuer). Fabien is working on the answers (Fabien
have also not to many time yet). I await the interview in the next days,
possible that Fabien will send today the answers. I am sure all what you say and
the others are thinking will be answered with the interview questions I create
for around two weeks.
April 03rd, 2005
CEGT or how many games are needed
93,
Heinz van Kempen
FQ:
For me is the most interesting still running tournament
the CEGT by Heinz van Kempen, Christian Koch and
Charles Smith. Reason enough to asked Heinz for more information, perhaps
for a little review?
Heinz sent it today in the evening ... GREAT!
Text by Heinz van Kempen
CEGT
or how many games are needed?
As a tester of chess engines for many years, I
always found it a bit frustrating that it was only possible to get a statistical
significant amount of games only at Blitz level when using not more than one
machine. Luckily, I could use three or four computers over the last months and
Christian Koch had the idea to combine my results with his on two faster
computers additionally and we are also getting support from Charles Smith in
order to run a very big event with a high amount of games and longer timecontrol
40/40 adapted to 2 Ghz P4 CPU via Crafty benchmark. At this time we have almost
1000 games each for most top engines.

We ran CEGT 1 with 510 games for each engine and
to be more up-to-date and to include more amateur engines we added CEGT 2 that
is also fairly advanced by now. Strikingly, we could again see that even with so
many games there are considerable differences, no matter if using Nunn positions
or general books. In the first tournament, Junior, for example was much better
than Fritz and Gandalf; clearly ahead of Hiarcs and now it is vice versa. SOS
started gorgeously but now dropped a lot and on the other hand Fruit now
performs much better.
For better understanding please compare the
overall cross tables for CEGT 1 and 2 respectively on our websites or have a
look at the rating progression calculated by Christian.
http://www.chessfighters.de/cegt/
http://www.husvankempen.de/nunn/
http://www.husvankempen.de/nunn/CEGTELO.xls
The conclusion we have to draw is that all
tournaments with less games and even with a few hundred games for each engine
only give accidental results. We will continue our experiment to see when we
will reach the point where ratings will finally be reliable and constant.
Besides of considering general weird statistics we want to have more insight
into what additionally can be concluded. So far we have strong indications that
there might be differences in performance on Pentium CPU´s on the one hand (used
by Christian exclusively) and on Athlon XP and Athlon64 CPU´s (used by Charles
and me). Please compare our individual overall scores and statistics and you
will see that for example SOS and Pharaon have better results so far on Pentium
and Chess Tiger and Deep Sjeng are doing better on Athlon´s. Will this change
with more games?
http://www.husvankempen.de/nunn/cegtstat.xls
Maybe we will continue this tournament for a long
time to give some reference. Anyway we will proceed as long as we have fun in
doing so and after all it is also interesting to watch those high level games.
March 31st, 2005
New interviews in
progress
92,
FQ
The computer chess community are never in
winter sleep. In this times we can produce the most news because many programmer
are working on his developements. Now the winter time ended and to search news for
Arena News-Ticker isn't easy. To produce news is easy because we have enough
interesting things to fill out daily a magazine.
At the moment four interviews are in progress:
All interviews will be available in
English language!
01. With Fabien Letousey (France)
to his developement Fruit. Robert Allgeuer (Luxembourg, his main
tester) sent much interesting material. The interview isn't ready. I believe the
next one you can read. At the moment I energized ... wait of his answers :-)
02. With Franck ZIBI (France) to
his development Pharaon. I work here in cooperation with Exacta and the
webmasters Christopher Conkie and Michael Diosi. The interview
will be public on Exacta webpage.
03. Middle up to end of april you can
read a bigger interview with one ot the top programmers. It's a little secret
and I work here in cooperation with Alexander Schmidt. The interview can
be found later on the webpage by Alex.
04. At the moment I am working on the
interview questions for Thinker programmer Lance Perkens (Canada).
Thinker is in my opinion one of the TOP 5 of the free available engines. Reason
enough to learn morre about the ideas from Lance. OK, I will give you some
interesting news to Thinker before the interview is available! Thinker will be
engine number 11 / 16 in my still running ATL-2 tournament. At the moment
Pharaon are playing the first games.


v. 4.7x
WB compatible
Home of
Thinker
Hint: Thinker is number 12 in ATL-1 Rating-List!
Text by Lance:
There is indeed a new version of Thinker coming
out in the next few weeks.
It includes:
- fixes to evalulation
- fixes to pondering
- fixes to time management for very fast ('bullet') games
- 64-bit version (for 64-bit Windows)
- a new tournament book
You will be informed if new
interviews are available.
See you ...
March 31st, 2005
SOS opening book available, v. 3.0
91,
FQ
About noon a new version of the
SOS opening book by Salvatore Spitaleri (Italy) is available.
Salvatore sent me much interesting comments I will public (with the permission
of Salvo).
Text by Salvatore:
I builded the book from my personal collection
of games, in every game a trap before the move number 35, so 70 halfmoves in my
book.
This is the output from "elopgn.exe" ,a
Windows(r) utility program written by Norm Pollock that statistically
analyzes elo values and games's results within a pgn file.
Utility by Norm Pollock can be find
under:
http://www.hoflink.com/~npollock/utility_files.html
number of games in collect.pgn is 49976
number of games with white elo is 49976
number of games with black elo is 49976
number of games with BOTH white & black elo is 49976
average white elo value is
2637.263186329438 for 49976 games
average black elo value is 2637.303545701937 for 49976 games
maximum elo distance is 297 for 49976 games
average elo distance is 63.82741716023691 for 49976 games
number of games with a result is 49976
number of games without a result is 0
number of white wins is 18768 (
37.55402593244757 % )
number of draws is 16383 ( 32.7817352329118 % )
number of black wins is 14825 ( 29.664238834640628 % )
white score over all games (with a result)
is 53.94489354890347 %
black score over all games (with a result) is 46.05510645109653 %
white score over non-drawn games (with a
result) is 55.8687821867651 %
black score over non-drawn games (with a result) is 44.1312178132349 %
Unfortunatly in Arena the construction of the
book isn't so easy, it is necessary to tune by hand the options, so I prefer
test my book in Chessbase, sorry.
Comment by Frank Quisinsky:
The ChessBase GUI have 10 years a healthy margin in developement as
commercial company. Much of the great features in ChessBase GUIs are also many
work by the programmers of ChessBase. More as one person are working on this GUI and many
Grandmaster and chess player gave in this time a lot of tips, hints and
impressions. We all know that the ChessBase GUI is one of the best but we all
know Martin Blume work too. For me Martin is the most talented GUI
programmer the World have (Martin don't want to hear this but live is hard). We are very content with the work on Arena. In a personal
meeting I discuss the book options with Martin and gave him much tips. Martin
changed different features but the newer version of Arena isn't ready for public.
Martin is still working on Arena.
After this comment I sent
Salvatore a mail and asked him for more details. Salvatore is one of the opening
book experts and his comments are very interesting for the others and of course
for us. Furthermore, we collect wishes by mail. Alexander Schmidt sent an
other bug report and after all I am sure the Gladiators will like the new
version of Arena.
Please, I want an alone change in book's option
to automatically generate my book:
minimal number of games ------> in percentage
minimal won games ------------> in percentage
Second comment by Salvo:
My first problem is in Book Option:
I would like:
minimal number of games ------> in percentage
minimal won games ------------> in percentage
to avoid moves that was played very rarely.
Otherwise I have to tune manually the rare moves assigning to these a
priority value of 0, (IMPORTANT) at every update of my book.
My second problem:
If I use a file with flag, the priority is
assigned by Arena with a bad formula, so Arena assigns "Priority 1" to "?" bad
moves, and not "Priority 0", so also "?" bad moves are played by Arena.
I suggest to assign priority:
0 to ?? Svista - Mossa molto cattiva -
Blunder - Very bad move
0 to ? Errore - Mossa cattiva - Bad move
0 to ?! Mossa dubbia - Dubious move
1 to Mossa senza annotazione - Without annotation move
5 to !? Mossa interessante - Interesting move
7 to ! Buona mossa - Good move
9 to !! Mossa molto buona - Very good move
like in Shredder Classic:
Movecomment Bookweights
!! - a
! - b
(none) or !? - c
?! - d
? - e
?? - f
and I would like a (NEW) Tournament option in
book (es. in tournament mode Arena will play only moves with a priority > 7 or 6
or 5... ; in Shredder Classic in tournament mode are played only moves
marked as a or b)
My third problem:
If I use 2 or more files with flagged moves to
build the book, the priority assigned to the moves of second file, overwrites
the priority that is already in the book, instead Arena would have to calculate
the average.
March 30th, 2005
Chess960, book by Sandro Necchi
90,
FQ
01.
Aice by Anastasios Milikas
(Greece) is since version 0.90 compatible to Chess960. This is number seven from
the group of available Chess960 engines. More and more engines will come in the
next times. This are great news for the Chess960 community and the organizers
for the first Chess960 WM in Mainz (Germany).
Aice:
http://www.milix.net/aice/
ChessTigers:
http://www.chesstigers.de
(Chess960 WM with a fantastic news page)
Smirf:
http://www.chessbox.de/Compu/fullchess1.html (information to Chess960)
Aice is a young and very popular
amateur engines. The programmer added UCI and WB support and from version to
version Aice is clearly improved. One of the young talents the amateur chess
area have. I am very happy that we have now ... after a long time ... a strong
engine from Greece. Unfortunately, we don't have a Greece translation from our
GUI but with time new translation will come.
02.
After my news to the opening book by Harry Schnapp I got some
positive and negative feedback. I know that many persons try to search in each
public news mistakes. The reason is easy! This group of self artist like
it to search a topic for a computer chess forum. A reason why chess fora aren't
very popular in times today. To many of this persons don't notice that the own
messages are the reason that others loosed the interest. This is a pity but we
have such situation not only in computer chess. This is a generally problem with
fora systems.

Without any words ...
Longley
:-) ... not shortly:
I have never wrote that the opening book by Sandro Necchi is bad. I like
the book by Sandro and used this book, the opening book by Dan Wulff and
under ChessBase 9.0 database the opening book from Grandmaster Boris Altermann (Junior team)
for my own interest. I am sure the book author from ChessTiger, Jeroen Noomen
and other authors, Sedat Canbaz, Denis Grafen, Prof. Dr.
Djordje Vidanovic, Andreas Runge, all the engine programmers and many
more try to make a good work too.
For the most isn't easy to understand that a new
developement can be good.
It's compare to: I don't know it, I can't
read about it ... it must be bad :-)
Unfortunately, such things we can
daily read. The press can be a reason for such opinions from person which need a
longer time for build an own one. This group of persons are the most interesting
for publications ... abide buyers! The German computer chess press build a
perfect example in the past!
Another one: Arena is freeware, oh a freeware
... of course this must be bad.
Always the same
for the hobby we like. Must thinking on the Ruffian discussion after the first
release version. The group of self artist are collect in an amusement
park, means computer chess fora with a grateful public.

The clone discuss to Ruffian for
around two years!
All have much fun with the duffy comments by known self artist in computer chess
fora.
Here the adept Crafty user club :-)
No opening book
will be perfect and all good books have problems in different of the openings. A
persons which have the opinion that only known book authors comes with a perfect
work must be a hobby player with very small chess
knowledge.
The same for the opening book from Harry Schnapp. This isn't a
perfect book but one of the best available opening books today. Our experts have
the opinion that it's possible that an opening book can make an engine 100-200
ELO stronger. Different thinking on 700 ELO!
Yes, this are the
experts today?
The same experts writes that the opening book from Harry isn't one of the
best available opening books. The reason is that after 400 games with longer
times controls Harry's book produce fantastic results.
A new competition
... shocked ... a new competition
An evidence / proof of
incapacity!
In my humbly
opinion:
We have to learn to understand, we have to learn that not only 20 persons are
working on engines and opening books and we have to learn to have a look for
other developments without to have own interest only in our brain. After this we
all can have much more fun with our hobby computer chess and with the work by
others. We have more possibilities and the slogan by the movie story of
Enterprise ... THE ENDLOSS WIDENESS OF GALAXY ... THATS computer chess!
With own
interest only you are very alone in chess fora today. Perhaps the reason that
some pesons try it with provocation if the own work don't find many comments by
others.
See news 89 or
others: Arena is not the only one of available interfaces.
And now ... let me look in Pirat :-)
The typical men's
World.
Let me give this one ... we need much more womens.
Then we all are peaceful friendly lambs.
I have more
interest to talk here about computer chess news ...
A lot of interesting news will come, you can be sure and each one with interest
to help to build a little online magazine are welcome.
March
30th, 2005
Pirat 1.01 GUI released
89,
FQ
Jürgen Wolf released Pirat
1.01 GUI and Pirat 0.3.3 Engine.
http://www.go-pirat.de/
On the webpage from Jürgen you can
find some screeshots and many additional support information about the
developement in German and English. I think its very good for the future that
many other programmers are working on a GUI for chess engines. Pirat is
compatible to UCI protocol 1 and 2, Autoplayer and have much more features.
Pirat is Sharware
The price after 3 months of
testing is EUR 29,95!
If you have question please
contact direct the programmer Jürgen Wolf.
I have no more information.
March
28th, 2005
Casa Lisa ... Arena and Cosmetic
88,
FQ
|
Lisa supported the
Arena development since around two years with some wonderful ideas. Lisa was working
for a long time in a
Cosmetic company as director. Today she is working mainly in distribution of cosmetic
ampules. We have some interesting offers to organize a computer chess event
with a very known cosmetic company. At the moment we don't have the time to make
these organizations and isn't the right time but the idea is just great.

http://www.casalisa.com
Lisa sell at wolesale products for womens.
Some of them with exclusive distributions and others are own productions.
My girlfriend liks consulting to the end customers. In the latest months
she created her own shop system. If you seach a nice present for your
wife or girlfriend, you can asked Lisa! I think this is a nice
additional service for our female parts. |
 Lisa and
myself
(Leiden 2003).
The Arena team won the cup with
Ruffian by Per-Ola Valfridsson (Sweden) |
In detail:
The connections we have will be used for an internet event after the release
version of Arena 2.0. This will be an online event with one of the
Gladiator-Shop engines powered by one of the biggest cosmetic companys in the
World. You can win nice prices and you will have much fun. In the past I had
organized for Millennium GmbH and Gambit-Soft "Deep Shredder vs. The World". I
will try to start a comparable event but much more interesting. Intel and a
German bank are the sponsers from this older event.
We will give you some more
details later ...
March
26th, 2005
Mainbook 5.0 by Harry Schnapp released
87,
FQ
Harry Schnapp released today a new
version from his very popular Arena Mainbook.
Arena Mainbook (detail page,
in English and German available)
This opening book isn't from a PGN file.
Harry Schnapp enter each variation in his book by hand. He is working over
one year on this opening book. Harry try to optimize his opening book for all
available chess engines. Critical gambit openings wouldn't add. This opening
book is really a sensation. Shredder 9.0 produced on my PCs fantastic results
with the book from Harry. If I compare the results with the opening book
by Sandro Necchi, the book by Harry produce an higher Shredder
performance (around 20 ELO). I will not say the book by
Sandro isn't good but many variants Shredder like are
deactivated.
Example:
ATL-2 results: Shredder is playing with Mainbook 4.5 by Harry Schnapp:
Rating 1: All games from
ATL-2 Systems 1 - 3 = 1460 games!
March 26th, 2005 (21:00 MEZ)
Program Elo + - Games Score Av.Op. Draws
1 Shredder 9.0 : 2750 29 29 420 71.9 % 2587 31.4 %
2 Gandalf 6.01 : 2645 28 27 420 56.2 % 2602 32.4 %
3 TheKing 3.33 Schumacher : 2621 28 28 400 51.7 % 2609 31.0 %
4 Ruffian 2.1.0 : 2606 29 29 380 48.6 % 2616 30.8 %
5 ProDeo 1.1 : 2605 29 29 380 48.4 % 2616 31.6 %
6 SlowChess Blitz WV : 2560 31 31 360 40.4 % 2628 28.1 %
7 SOS 5.1 for Arena : 2547 30 30 360 38.2 % 2630 32.5 %
8 Spike 0.9 : 2530 40 41 200 31.8 % 2663 33.5 %
After 420 games the advantage to
Gandalf 6.01 = 105 ELO !!
This games are produced with 40 moves in 20 minutes (still running ATL-2
tourney).
ATL League (more information, games, log
files and so on)
On my secret ATL-2a tournament (Athlon 2.8,
Centrino 1.6) with Shredder Classic GUI the different from Shredder 9.0 to
Gandalf 6.01 is 82 ELO. After 340 games for each engine with the same time control
40 moves in 20 minutes.
Shredder is playing with the opening book by Sandro Necchi. I will public
this tournament later. Perhaps I produce other results in tournaments with Fritz,
Hiarcs, Tiger and Junior and perhaps the opening book by Sandro is tuned against
this group of engines. Or simply, around 400 games are not enough.
FACT is:
The opening book by Harry Schnapp is really a sensation and one of the
best available opening books today!
Update text by Harry Schnapp in German
language:
Von Harry Schnapp:
Ein einheitliches Buch das in der Lage sein muss bis zu 250 Engines zu bedienen,
hat in seiner Erstellung eine Reihe von Hindernisse zu bewältigen. Darunter
nicht nur möglichst ausgewogene Stellungen, sondern auch solche, die nicht
eindeutig eine bestimmte Spielart favorisieren. Um dieses Ziel zu erreichen,
sollten gefährliche Gambiteröffnungen vermieden werden, wie auch übertrieben
lange Zugfolgen. Aus den Erfahrungen der vorherigen Versionen habe ich
festgestellt, dass besonders Schachprogramme mit "mittelmäßigem Level" oft nicht
in der Lage sind, ein Materialopfer auszunutzen oder beispielsweise sich richtig
in eine offene scharfe Stellung zu orientieren, wenn die entsprechende Engine
"x" auf ein strikt positionelles Spiel aufgebaut ist. Daher lieber manchmal die
langen Varianten mehr oder weniger kürzen, um den NEUTRALEN Punkt ermöglichen zu
können. Ab hier sollten die Programme, rechnerisch, selbst die für ihre
Spielweise richtige Fortsetzung finden. Auch starke Engines würden überlastet
sein, wenn die Stellung beim Verlassen des Buches nicht passt. Unlängst habe ich
aus einem Match Shredder 9 gegen Spike 0.9 einige sehr interessante Partien
beobachtet. Shredder spielte mit dem Mainbook 4.5 und Spike mit dem wirklich
guten Buch von Frank Quisinsky. Es passierte folgendes: In einer Partie hat
Spike den Shredder in einer gekürzten Variante "erwischt" und war daher mit 4
Züge im voraus. Der Zeitvorsprung (Shredder musste ja rechnen während Spike aus
dem Speicher spielte) verpuffte langsam und letztlich gewann Shredder problemlos
diese Partie. Grund: die Variante war zwar sehr lang, passte aber nicht gerade
zu Spike. Dagegen in andere vier Partien aus dem selben Match wo diesmal
Shredder um einen Zug die Nase vorn hatte, holte Spike einen Sieg, ein Remis und
verlor zweimal nur sehr knapp im Endspiel! Grund: Mit diesen Varianten kam Spike
gut zurecht und konnte dem übermächtigen Shredder 9 Paroli bieten. Eigentlich
hatte ich die Absicht meine neue Version Mainbook 5.0 nur leicht gegenüber
Version 4.5 zu ändern, aber im Laufe der Zeit fand ich sehr interessante gut
analysierte "Mittelstreckler" die in meinem Konzept passten und die habe ich
eingebaut. Es sind Varianten mit Zügen zugefügt worden aber auch 83 die ich
leicht kürzen oder ändern musste aus den oben genannten Gründen. Wer das Mainbook 4.5 hat, sollte es eventuell auf der Festplatte behalten, besonders als
Engine Buch für den Shredder 9. Ob die neue Version 5.0 genauso geeignet ist,
für diese Super-Engine von Stefan Meyer-Kahlen, das wird sich im Laufe der Zeit zeigen.
Möglich ist alles!
Ich möchte nochmals darauf
hinweisen:
Harry ist über 70 Jahre alt und zeigt uns Jüngeren so richtig schön was möglich
ist.
Ich wünsche mir, dass ich im Alter von 70 Jahren mit dieser Energie unser Hobby
nutzen kann.
Read more on under:
Arena Mainbook (detail page,
in English and German available)
Arena Mainbook:
158.848 variations, 1.780.651 moves, depth = 63 plys!
Mainbook
v. 5.0 (5,72 Mb, RAR format)
Update Historie:
Version 3.0: Increased by 18.734 variations and 318.269 moves (version
3.0 isn't longer available).
Version 4.0: Increased by 11.994 variations and 223.456 moves (version 4.0 isn't longer available).
Version 4.5: Increased by 01.003 variations and 018.238 moves.
Version 5.0: Increased by 14.279 variations and 227.714 moves.
Zu den neueren Versionen
schreibt Harry (German text):
Version 4.0: Allrounder bei dem alle Engines frei, ohne Begrenzung
spielen können mit einfacheren üblichen Positionen.
Version 4.5: Es ist ein Update von zusätzlich 1.003 Varianten mit 18.238
Zügen, sehr geeignet für stärkere Engines die problemlos ohne Limit spielen
können. Für die Mittelklasse und kleinere Programme sollte ein Limit von 40
Halbzüge gesetzt werden um bei sehr langen Varianten strategisch anspruchsvolle
Stellungen zu vermeiden.
Version 5.0: Da keine PGN Dateien mit Partiesammlungen verwendet wurden,
sondern analysierte Eröffnungen aus der Fachliteratur, wurde wie immer, die
Bezeichnung Varianten und nicht Partien verwendet. Weitere Informationen finden
sich in News-Ticker Mitteilung 87!
March
23rd, 2005
Timing with a neural network
86,
FQ
Volker Annuss released
today Hermann 1.31.
UCI / WB / FRC
Hermann
Volker made in
the last weeks a lot of experiments to an interesting topic.
He worte a review for Arena webpage.
Timing with a neural network
by Volker Annuss
For a chess engine it is difficult to decide
how much time to use for one move. In some positions one single move, for
example capturing a queen, seems to be the best. Nevertheless it is risky to
spend only a little time on such a position and take the queen immediately.
Another approach is to spend more time on
positions that are not clear.
Hermann uses a neural network to identify these positions.
How it works
For an engine that uses iterative deepening, a
position is not clear (by definition), when it is likely that the best move will
change, when iterating one ply deeper.
Hermann uses a neural network to calculate a
probability of how likely it is, that the best move will change when iterating
one ply deeper. This probability is calculated from node counts gathered during
the search process.
Only four input values are used:
- the percentage of nodes visited for the best move.
- the percentage of nodes visited for the three other moves with the highest
node counts.
Node counts from many games have been recorded
together with the information whether the best move has changed the next
iteration or not. Then the neural network has been trained with this data.
Now when Hermann searches a move, after every
iteration the neural network predicts a probability of how likely it is that the
best move will change, when iterating one ply deeper. Depending on this
probability, some additional time is spent on this move.
Verification
The aproach has been verified in some Chess960
games. Hermann 1.3 Neuro has timing with neural network enabled. Hermann 1.3
Classic has it disabled. His timing works like the versions before 1.2.
Test 1:
15 minutes for the whole game on a Pentium IV 2667 MHz
Hermann 1.3 Neuro : Frenzee
2.00 3.0/20
Hermann 1.3 Classic : Frenzee 2.00 4.0/20
Hermann 1.3 Neuro : Chispa
4.03 10.0/20
Hermann 1.3 Classic : Chispa 4.03 4.0/20
Hermann 1.3 Neuro : The Baron
1.0.5 4.0/20
Hermann 1.3 Classic : The Baron 1.0.5 4.0/20
Hermann 1.3 Neuro :
all 17.0/60
Hermann 1.3 Classic : all 12.0/60
Hermann 1.3 Neuro : Hermann 1.3
Classic 12.0/20
Test 2:
2 minutes plus 6 seconds for every move on an Athlon64
3400+
Hermann 1.3 Neuro : Frenzee
2.00 5.5/30
Hermann 1.3 Classic : Frenzee 2.00 4.0/30
Hermann 1.3 Neuro : Chispa
4.03 14.5/30
Hermann 1.3 Classic : Chispa 4.03 13.0/30
Hermann 1.3 Neuro : The Baron
1.0.5 10.5/30
Hermann 1.3 Classic : The Baron 1.0.5 7.5/30
Hermann 1.3 Neuro :
all 30.5/90
Hermann 1.3 Classic : all 24.5/90
Hermann 1.3 Neuro : Hermann 1.3
Classic 16.0/30
The difference between these versions is not
always that big. It is much smaller in normal chess games where the first moves
are played from an opening book, so most of the improvement might come from more
time that is used in the opening.
March
22nd, 2005
New files on Arena
85,
FQ
In the time I am
ill I added some files in "User Files, Engines" selection. Our News-Ticker and
the "User Files, Engines" selection are the most interactive parts on Arena
webpage. Daily updates can be found on ATL-League.
- Morton Skarstad (Norway)
sent the Norwegian translation for Arena 1.1. The menu system of Arena is
now available in 18 languages. Again a new record! We hope that we get some
translation more. Popular languages, like Greece, Danish, Finnish and much
others are not translated yet. Please send me a mail if you have interest to
help us with a translation!
- Today Sedat Canbaz (Turkey)
sent a new version from his Perfect opening book. Our book expert
Harry Schnapp have a very good opinion from the work by Sedat and I like the
preview version too. It seems that he spent a lot of time working on his opening
book.
Webpage by Sedat
Canbaz
- I found a new version of
CYGWin1.dll and added the file a view days ago. You need the file for
different engines compiled / develops under Linux operating system (good
examples are the engines Gothmog and Phalanx).
- Gabriel LEPERLIER
(France) updated his opening book for a while too. The opening book from Gabriel
have a nice idea. Gabriel used games from his own eng-eng tournaments only.
Please visiting
from time to time the "User Files, Engines" selection.
March
21st, 2005
Ktulu 7.0 release information
84,
FQ
Included in the interview with
Rahman Paidar (Iran), I added for some days, you can find a lot of
information to the developement of Ktulu. After around one year it's time for a
new version of Ktulu and Rahman and myself discuss the prices again.

Ktulu 7.0 will be
available around
April 15th, 2005
Customers of Ktulu 5.0 / 5.1 have
to pay for Ktulu 7.0 EUR 14,95.
New customers have to play EUR 19,95.
If you have more questions you can
send me a mail.
sales@nospamgladiatorshop.com (enter the mail
address with "nospam").
If I have more information you can
find it _later_ under
Gladiator-Shop and of
course in Arena News-Ticker.
Ktulu 7.0 will be start in ATL-2 League soon (after Spike 0.9 as engine
number 9).
March
21st, 2005
SOS and hashtables
83,
FQ
The topic are the different
SOS.5 for Arena results on ATL-2 League. After around 40 days 1.200 games
with 40 moves in 20 minutes are played. On Dual Xeon 2.8 GHz with 256Mb
hashtables SOS.5 for Arena seems to be 150 ELO stronger compare to Athlon64 3.8
GHz with 128Mb for hashtables. I asked Rudolf Huber for more information
(perhaps SOS is optimize for Intel processors) and he sent me an interesting
comment in German:
Go and look:
ATL League

Rudolf Huber (programmer of
SOS):
Sehr interessant das Ganze. Ich habe eine andere Vermutung dazu:
SOS braucht sehr viel Hash. Ca. 10 MB pro Sekunde Bedenkzeit. Auf dem XEON
bekommt die Engine am meisten Hash und spielt stärker. SOS ist zwar sicher auf
dem Athlon 64 3.8 viel schneller als auf dem XEON, aber mit nur 128 MB Hash wird
die Suche dort zu stark gebremst. SOS ist sicher nicht für Intel optimiert. Ich
habe ja selbst einen Athlon64.
I try to
translate in English:
Very interesting at all. I have an
other guess to that:
SOS need many hashtables. Around 10Mb per second time limit. On Xeon SOS
got the most hash in ATL (256Mb) and will be playing stronger. Indeed on Athlon
64 3.8 GHz SOS is faster compare to Xeon 2.8 GHz but with only 128Mb hashtables
the search will be slowed down (limited). For sure, SOS isn't optimize for Intel
processors. I am using by myself an Athlon 64 processor.
Example:
10Mb hashtables per second time limit!
I am playing in ATL with 40 moves in 20 minutes = around 30 seconds per move.
30 x 10Mb = 300 Mb for hashtables ... and I am using 128 Mb hashtables only!
March
21st, 2005
Shredder's users manual in PDF
82,
FQ
In the latest two
weeks I am ill. I made only the "important" updates on Arena webpage. I will try
to hold the News-Ticker again up to date with a lot of additional information to
our hobby computer chess.
Shredder / Achilles programmer
Stefan Meyer-Kahlen added a user manual in PDF format for his software on his
own webpage. Again a very good service for his customers. Today the PDF format
is standard.
Note:
Since a while a
new Adobe Reader version is available!
Version 7.0 can be found under:
http://www.adobe.com/products/acrobat/readstep2.html?c=us
Adobe Reader is the program you
need for reading PDF files but ...
Very interesting in working with
PDF files is the free Open Office packet!
Version 1.14 can be found under:
http://www.openoffice.org/
With modern browsers you can load
PDF files directly.
Very interesting are
FireFox 1.01
and
Avant Browser
Additional information to FireFox (German language)
can be read here:
http://firebird-browser.de/
Enough interesting downloads for
the evening, you will have a nice time of testing and of course thanks Stefan
for the user manual in PDF format.
March
20th, 2005
Spike 0.9 released
German interview with
Volker Böhm and Ralf Schäfer
81,
FQ


v. 0.9
UCI / WB compatible
Home of
Spike (Arena Partner)
Interview with Volker Böhm in cooperation with Ralf Schäfer
Hint: Spike 0.9 X2 is number 25 in ATL-1 Rating-List!
Frank Quisinsky:
Zunächst möchte ich Euch zu den Erfolgen beim CCT-7 und dem IPCCC 2005
gratulieren. Sind wir ehrlich, mit den Ergebnissen konnten wir gar rechnen! In
der ATL-1 war schon ersichtlich, dass Spike 0.9x um mehr als 100 ELO zugelegt
hat. Dies gibt mir natürlich reichlich Material um nachzufragen. Wie kam es zu
diesem deutlichen Sprung? Wurden Teile von Spike komplett umgeschrieben? Welche
Änderungen waren für diese enormen Spielstärkezuwachs verantwortlich?
Spike Team:
Für Spike 0.9 haben wir uns etwas mehr Zeit gelassen. Wir denken, dass weniger
häufige Releases mit größeren Spielstärken Verbesserungen auch bei den Anwendern
und Testern bzw. Turnierveranstaltern besser ankommen. Durch die zusätzliche
Zeit konnten wir auch mehr Ideen implementieren. Im Wesentlichen haben wir
Verbesserungen in der Stellungsbewertung gefunden, insbesondere Mobilität und
verbesserte Freibauern. Außerdem haben wir mit History-Pruning experimentiert
und eine Lösung implementiert, die gut zu Spike passt. Ein paar weitere
Verbesserungen hier und da haben dann das Ergebnis gebracht.
Frank Quisinsky:
Eine so deutliche Programmverbesserung steigert natürlich auch die
Erwartungshaltung der Fans. Natürlich wird es nicht möglich sein, Spike beständig
in 100 ELO Schritten zu verbessern. Unzweifelhaft ein riesiges
Potential zu erkennen. Ich gehe davon aus, dass Ihr derzeit die Prio auf
Spielstärkeverbesserung legt und auch in den nächsten Monaten sich nichts daran
ändern wird. Dennoch wäre es interessant zu wissen, ob Ihr auch hinsichtlich
Kompatibilitäten Spike in absehbarer Zeit verbessern möchtet. UCI II ist ein
riesiges Thema und auch der Support zu Chess960 ist in aller Munde. Wird Spike
bei der ersten Chess960 in Mainz an den Start gehen?
Spike Team:
Mit der 0.9 haben wir schon ein bisschen in Kompatibilität investiert. Z.B. sind
einige Teile von UCI II integriert. Die "Refutation"-Implementierung gefällt uns
aber noch nicht, da sie zumeist nur die PV ausgibt. Außerdem haben wir unser
Interface zur Fritz GUI angepasst. Neu ist auch die Möglichkeit einige
Einstellungen für Search konfigurieren zu können. Chess960 werden wir in der 1.0
haben. Wir freuen uns schon auf das Turnier in Mainz, bei dem wir gerne
teilnehmen!
|
Frank Quisinsky:
Es ist heute gar nicht mehr so ungewöhnlich, dass sich zwei Personen bei einer
Programmentwicklung vereinen. Natürlich wird eine solche Kooperation gerade hinsichtlich der
Programmierung kompliziert, denn eine Engine bezeichne ich mal als
individuellen Ausdruck eigener Ideen. Dennoch spricht alles dafür, sich in einem
"kleinen" Team auszutauschen, mit dem Ziel, die freundschaftliche Verbindung mit
einem gemeinsamen Interesse zu vereinen. Der bisherige Erfolg spricht eindeutig
dafür, dass der Zusammenschluss Früchte trägt. Wie kam es zu Euren
Zusammenschluss und seit wann arbeitet Ihr in einem Team an Spike?
Spike Team:
Wir haben beide zu ähnlichen Zeiten angefangen ein Schachprogramm zu erstellen.
Das ist dann schnell eingeschlafen, da die Vergleichsprogramme (Kommerzielle)
einfach viel zu gut waren. Vor etwa zwei Jahren hat Ralf dann Arena entdeckt und
damit die Möglichkeit auch gegen einfache Programme zu spielen. Das gab uns die
Möglichkeit schnell Gegner zu finden, gegen die wir auch gewinnen können. Am
Anfang haben wir beide unterschiedliche Engines entwickelt, Cheetah und IceSpell.
Von Anfang an haben wir aber über die Ideen darin diskutiert. Zumeist hat Volker
über die komplizierten Algorithmen gegrübelt und Performance optimiert, Ralf hat
mehr an Bewertungen gebastelt. Da kam dann die Idee auf uns einfach
zusammenzuschließen. Jetzt macht Ralf die Bewertungen und Volker die
Datenstrukturen und die Suche. Dadurch kommen wir uns nicht ins Gehege. Nach wie
vor werden aber die meisten Ideen gemeinsam diskutiert und getestet. Spike ist
inzwischen ein Jahr Alt - seit der ersten Programmzeile gerechnet.
|

Ralf Schäfer (links) und Volker
Böhm (Spike),
hier mit Achilles / Shredder Programmierer
Stefan Meyer-Kahlen (im Vordergrund rechts)
Aufnahme beim IPCCC 2005 in Paderborn! |
Frank Quisinsky:
Trotz der vielen Programmentwicklungen bzw. der Vielfalt an zur Verfügung
stehenden Webseiten der Engine Programmierer, fällt Eure Webseite besonders auf.
Da steckt sehr viel Liebe im Detail und lässt vermuten, dass ein weiblicher Part
im Hintergrund Fäden zieht. Ist das nur eine Vermutung oder möchtet Ihr
auspacken? Ach ja, die eigene Erfahrung lässt grüßen ...
Spike Team:
Na ja, wir haben hier professionelle Unterstützung. Die Lebensgefährtin von Ralf
erstellt beruflich Web-Seiten und hat uns auch bei den Seiten von Spike ein
schönes Layout erstellt.
*** Achtung: Spike 0.9 startete
heute in die ATL-2 ***
ATL League
Frank Quisinsky:
Es ist erstaunlich das Programme wie Spike mit Verspätung unsere Festplatten
beglücken. Habt Ihr später das Internet als Plattform entdeckt bzw. seit erst
später in den Genuss der Informationen aus dem WWW gekommen? Waren andere
Entwicklungen aus dem WWW verantwortlich, für die offenbar nunmehr verstärkte
Entwicklung von Spike?
Spike Team:
Nein, Spike ist einfach sehr jung, d.h. inzwischen genau ein Jahr alt. Wir
wollten es erst freigeben, wenn wir eine Engine haben, die zumindest
einigermaßen spielt. Das war dann die Version 0.7, die immer noch ab und zu
runtergeladen wird.
Frank Quisinsky:
Turniere wie das IPCCC in Paderborn sind leider nur bei einer kleinen Gruppe von
Engine Programmierer beliebt. Das hat mehrere Gründe wie z. B. Zeit, berufliche
Verpflichtungen oder die Anreise aus fernen Ländern. Viele der teilnehmenden
Programmierer vom IPCCC sind Dauergäste und kennen die Lokalitäten im Umfeld von
Paderborn. Wie seit Ihr als "Newcomer" außerhalb des Turniersaals aufgenommen
worden? Es waren sicherlich viele aus den Foren bekannte Personen beim IPCCC und
haben Euch Löcher in den Bauch gefragt :-) Oder wart Ihr
über den Zuspruch aus der Usergemeide eher enttäuscht?
Spike Team:
Als wir angekommen sind waren wir erst etwas zurückhaltend empfangen worden.
Einige wenige Teams saßen hinter ihren Rechnern und haben irgendwas an ihren
Systemen geschafft. Dann aber, insbesondere während der Turniere gab es dort
eine tolle Atmosphäre und viele schöne Gespräche mit anderen Entwicklern. Aber
Löcher im Bauch hatten wir nach IPCCC eigentlich nicht.
Frank Quisinsky:
Irgendwann muss ich ja etwas dazu schreiben?! Das Spike Team ist uns sehr
sympathisch und wir fragten Euch nach einer Partnerschaft mit uns. Diese
Partnerschaften sind für uns sehr wichtig. Einigkeit im Amateurschachbereich und
gegenseitige Achtung von erbrachten Leistungen ist das Resultat von dem was wir
heute vorfinden. Zu animieren wird aufgrund vieler selbsternannter Experten und
Selbstdarsteller immer schwieriger aber auch das zeugt davon, dass die Gruppe an
interessierten Anwendern zumindest größer wird. Wir freuen uns sehr über Eure
Zustimmung und sehen einer langfristig ausgelegten Partnerschaft entgegen. Spike
wird viele Fans finden und wir hoffen natürlich auch, dass die Gruppe der
Anwender der Arena Chess GUI größer wird. Was sind Eure größten Wünsche
im Hinblick auf die laufende Arena Entwicklung?
Spike Team:
Arena ist ein tolles Programm und super umfangreich. An Funktionalität fehlt uns
eigentlich nicht viel. Im Turnier bei IPCCC haben wir mit Arena gearbeitet und
für die Präsenztermine fehlten uns ein paar Kleinigkeiten - eventuell wussten
wir auch nicht wie das geht: Es fehlte uns eine Möglichkeit die Arena-Uhr an die
reale Uhr anzupassen; es gibt keine Schnelleingabe von Zügen (z.B. bei
Eindeutigkeit einfach ein Klick auf das Zielfeld) und wir hatten keine
Möglichkeit gefunden vor Beginn des Spieles das Spiel so zu beschreiben, dass
während des Spieles alle Einträge in der laufend geschriebenen PGN richtig
waren. Am meisten würden wir uns aber über das Korrigieren der kleinen Fehler in
Arena freuen, z. B im ICS-Modus. Eine Sache für Schachprogrammierer: eine
Möglichkeit über verschiedene Rechner laufende Turniere auf einem Rechner
zentral zu verwalten. Und vielleicht noch ein kleiner Wunsch, der uns
Programmierern das Analysieren von Problemstellungen erleichtern würde: Die
Möglichkeit, aus dem Variantenbrett heraus eine Fen in die Zwischenablage
kopieren zu können, so wie es auf dem normalen Brett ja auch schon geht.
Frank Quisinsky (im
Auftrag von Arena unterwegs):
1. Arena Uhr:
Roland Pfister (Patzer) experimentiert derzeit mit Chess960 Support. Im Rahmen
seiner Experimente hat er uns auch schon auf diesem Tatbestand hingewiesen. Die
Wünsche wurden an Martin Blume weitergeleitet.
2. Schnelleingabe von
Zügen:
In News-Ticker Message 27 gibt es einen Hinweis zu Arena 1.2. Das ist eine Super
Alpha Version über die ich hier berichtet habe. In dieser Version ist es möglich
die Schnelleingabe von Zügen auch mittels Tastatur durchzuführen. Die Idee mit
dem Klick auf das Zielfeld wäre natürlich eine weitere sinnvolle Ergänzung.
3. ICS-Modus
Wir haben fleißig über ein Jahr Vorschläge und Fehler zum ICS Mode gesammelt.
Martin wird keine wesentlichen Neuerungen beim ICS Support bringen aber die
Korrektur der Fehler hat eine sehr hohe Priorität. Es ist mühsam bei der
Kontrolle der erfolgten Bugs neu zu kompilieren um sich dann auf dem Server
anzumelden und zu prüfen ob die Korrekturen auch erfolgreich waren. Das kostet
Energie und ganz sicher ... dafür muss Martin ein paar gute Tage mit sehr viel
Motivation finden.
4. Laufende Turniere von
mehreren Rechnern auf einem Rechner zentral verwalten:
Also ich nutze mein Netzwerk für solche Operationen und greife direkt in die
*.at Datei ein. Ich wüsste jetzt ehrlich gesagt auch nicht wie eine andere /
bessere Idee aussehen könnte. Martin wird das Interview auch lesen und
vielleicht können wir Ideen per E-Mail austauschen.
Zusatz (German):
Volker Böhm berichtete am 21.03.2005 per E-Mail über eine Idee, die ich an
Martin weitergeleitet habe.
Steuerung von Turnieren von einem Rechner bei Verwendung von mehreren
Systemen!
5. Variantenbrett
Sehr schöner Vorschlag, wird Martin sicherlich schnell umsetzen können!
Frank Quisinsky:
Ich möchte in dem Interview mit Euch ein paar Meinungen zu User-Standard-Fragen
einholen. Wir werden ganz sicher noch öfters über die Spike Entwicklung
berichten und von daher habe ich breites Spektrum für dieses Interview, denn ich
habe die Möglichkeit öfters ein kleines Interview mit Euch zu veröffentlichen.
Ein Thema welches mich persönlich und derzeit stark beschäftigt sind
Eröffnungsbücher. Wenn ich mir das Buch von Harry Schnapp anschaue bin ich
schlichtweg begeistert, denn z. B. erzielt Shredder 9.0 bessere Resultate mit
dem
Buch von Harry als mit dem Buch von *Sandro Necchi*. Ich möchte nicht behaupten
das das Buch von Sandro nicht gut ist (ganz im Gegenteil) aber ich vermute, es
liegt auch an der enormen Spielstärke von Shredder. Übrigens, ein sehenswerte
Partie die Spike gegen Shredder in Paderborn gespielt hat. Viele Engine
Programmierer sind der Meinung, das ein gutes Buch von 200-700 ELO ausmachen
könnte. Ich halte die Zahlen für stark übertrieben. Seit Ihr mit Eurem Buch
zufrieden und wie hoch ist die Prio, dass eigene Spike Buch zu verbessern? Würdet
Ihr den Part des Eröffnungsbuches lieber auslagern oder einer dritten Person
übergeben?
Spike Team:
Man kann mit einem schlechten Buch eine Engine sicherlich 700 ELO schlechter
machen. :-) Eine gute Engine kann auch die Eröffnung ohne Buch sehr gut spielen.
Ein Buch muss daher schon einige Qualitäten haben, damit es besser spielt als
eine gute Engine. Hydra z.B. begrenzt das Buch auf wenige Züge obwohl sie sehr
kompetente Schachspieler im Team haben. Wir haben selber nicht das Schachwissen,
um ein Buch selber zu schreiben. Derzeit ist das Spike Buch komplett generiert.
Spike ändert sich noch zu schnell, als dass ein angepasstes Buch Sinn macht. In
IPCCC hat uns das generierte Buch nie in eine schlechte Stellung gebracht. Wir
waren sehr zufrieden. Wir haben schon ein Buch von einer dritten Person für
Spike bekommen, dass passte dann aber nicht mehr so gut zur aktuellen Spike
Version. Wenn wir weniger schnell neue Ideen für Spike umsetzen, dann freuen wir
uns sicherlich über Hilfe durch kompetente Buchschreiber. Spike hat übrigens ein
eigenes Programm (mit Oberfläche) mit dem man Bücher erstellen und manuell
nachkorrigieren kann. An dem Programm wollen wir noch irgendwann
weiterentwickeln, so dass wir die Einträge im Buch auch mit Spike überprüfen
können.
Frank Quisinsky:
Tuning ist ein heißes Thema. Ob nun Engines mittels Overclocking vom Prozessor /
RAM getunt werden oder versucht wird zu beeinflussen Engines mittels 5-Steiner,
6-Steiner Nalimov Table-Bases, hohe Hashtaballen oder Beeinflussung der
Parameter zu Höchstleistungen zu animieren. Der User nutzt jede Chance um im
stillen Kämmerlein aus einem Spike ein Spike / Müller oder Spike / Meyer zu
fertigen. Ich erinnere an TheKing und die vielen Einstellungen die möglich sind
bzw. den vielen Fans solcher Parametereinstellungen. Wie dem auch ist, eine
Engine ist zunächst mal Chefsache und der Programmierer fährt die Fahrwerke aus
bzw. steuert den Flieger auf ein Ziel zu. Bevorzugt Ihr ein Navigationssystem in
Kombination mit dem User oder ohne den User? Damit meine ich mögliche Parameter
für Spike. Es ist meines Erachtens erst dann sinnvoll viele Parameter zu
schaffen, wenn eine gewisse Spielstärke erreicht ist bzw. die wichtigsten eigenen
Ideen umgesetzt sind. Erst dann sollten Parameter Veränderungen wirklich für
eine deutliche Beeinflussung der Engine sorgen. Strebt Ihr zusammen mit den
Usern an die Engine zu verbessern bzw. möchtest Ihr langfristig ein
zuverlässiges Test-Team bilden um gezielt und mit Unterstützung des Users auf
Fehlersuche zu gehen bzw. Spike weiter deutlich zu verbessern?
Spike Team:
Spike hat mit der 0.9-er einige Parameter für Search. Wir beobachten mal, was
dabei rauskommt. Insgesamt haben wir aber den Eindruck, dass viele Tester so
wenig Spiele machen, dass wir es statistisch nicht für relevant halten. Aktuell
machen wir für den Test einer neuen Einstellung ca. tausend Spiele. Wenn jemand
Interesse hat verschiedene Konfigurationen in dieser Größenordnung unter
sauberen Konditionen zu testen dann können wir uns auch hier Unterstützung
vorstellen.
Frank Quisinsky:
Unter den Lesern von diesem Interview sind sicherlich auch viele Programmierer.
Die Programmierer interessieren sich wenn überhaupt eher für die
programmiertechnischen Ideen. Welche Standards habt Ihr in Spike implementiert?
Was ist das Geheimnis der hohen Spielstärke von Spike und welche langfristigen
Ideen könnten dieses Spielstärke von Spike zu weiteren größeren Sprüngen
treiben? Wo seht Ihr das Geheimnis von Engines wie Shredder oder Gandalf?
Shredder dreht derzeit einsam seine Runden, liegt fast 100 ELO vor der gesamten
Konkurrenz und das ist mehr als nur ein Klassenunterschied. Vermutlich sind
viele Profis zutiefst deprimiert wenn beobachtet wird, wie Shredder die gesamte
Konkurrenz an die Wand spielt. Es muss doch möglich sein Stefan Meyer-Kahlen
den Fehdehandschuh erfolgreich vor seinem Source Code zu werfen? Denkt Ihr das
auf absehbarer Zeit ein Amateur das Spielstärken Niveau von Engines wie Hiarcs,
Gandalf oder TheKing erreichen wird oder das gar ein Amateur an dem Thron von
Stefen sägen könnte?
Spike Team:
Wenn wir mal 100 ELO stärker spielen als die dann aktuelle Version von Shredder
sind wir sicherlich auch kommerziell. Im Ernst, das Shredder Niveau erreicht man
nicht nur mit guten Ideen, man muss auch sehr viel Zeit aufwenden. Die vielen
umfangreich getesteten Details sind aus unserer Sicht eines der "Geheimnisse"
von Shredder. Als Amateur werden wir schon aus Zeitgründen nicht auf
Shredder Niveau kommen. Spike ist aus unserer Sicht auch nur so gut geworden,
weil wir unsere Ideen gut testen. Zum Einen haben wir Debuggingcode in Spike
integriert, der uns eine sehr gute Analyse des Suchbaums ermöglicht und zum
Anderen testen wir neue Ideen mit vielen Spielen. Natürlich spielen auch
Begeisterung und gute Zusammenarbeit eine Rolle. Wir beiden ergänzen uns
hervorragend.
Frank Quisinsky:
Der Naum Programmierer *Aleksandar Naumov* gab einem interessanten Kommentar in
einem Interview mit mir ab: "Engine info you ask about is interesting, but I
would also like to know more about the person (age, birth place, profession,
education,...)." Ich bin mir sicher, dass Ihr Euch nicht lumpen lasst und die
gewünschten Informationen weitergebt :-)
Spike Team:
Wir sind beide 1967 geboren, Informatiker seit dem Studium an der Uni in
Kaiserslautern und wohnen im Raum Wiesbaden - wo wir auch gemeinsam zur Schule
gegangen sind. Ralf ist Entwickler, Volker ist Projektleiter in einem IT-
Unternehmen. Noch eine Gemeinsamkeit: wir haben beide sehr tolerante
Lebensgefährtinnen, die uns neben der Arbeitszeit auch noch für die
Schachprogrammierung entbehren.
Frank Quisinsky:
Ich möchte Euch nicht zu viele Fragen in die Interviews mit den Engine
Programmierer bringen oder mir gehen auf Dauer die Fragen aus :-) Aber diese
letzte Frage sei mir gestattet ... Wird Spike in absehbarer Zeit ein bedeutendes
Turnier gewinnen? Erlaubt ist nur ein Ja oder Nein :-)
Spike Team:
Nein
Frank Quisinsky:
Vielen Dank für die Bereitschaft zum Interview! Wir alle wünschen auch Euch
weiterhin viel Glück und Erfolg mit Spike! Wann wird Spike 0.9 (warum nicht
1.0?) verfügbar sein. Die Usergemeinschaft wartet schon sehnsüchtig auf die
deutlich verbesserte Version. Ganz ehrlich, ich bin auch ein kleiner Raffzahn
:-)
Spike Team:
In ganz kurz. Die Version ist fertig, wir müssen sie nur noch herausbringen.
Spike hat mit 0.1 angefangen, dann gab es 0.2, ... Jetzt sind wir halt bei 0.9,
danach kommt die 1.0. Wir wollen für die 1.0 eine vollständige Chess-Engine,
spielstark und mit allem was man so braucht. Die 0.9 ist fast so weit - aber nur
fast. Es fehlt zum Beispiel EGTB Support.
March
15th, 2005
Ktulu 6.2 will be released soon
Interview with Rahman Paidar
80,
FQ


v. 7.0
UCI / WB compatible
Home of
Ktulu
Gladiator-Shop
Interview with Rahman Paidar by Frank Quisinsky
Hint: Ktulu is number 13 in ATL-1 Rating-List!
Frank Quisinsky:
We had a long time no intensive contacts. I believe
you are very busy and maybe you are working on your "career" or better
occupational qualification / professional qualification? Sorry, I don't find the
right words but I am sure you know what I mean.
Rahman Paidar:
Well, I am still studying at university. It is the last semester and I'm also
working in parallel on my final university project.
Frank Quisinsky:
Are you working in your free time with the same interest as in the year
before on Ktulu?
You are one of the young talents we have and we all wish you a
long time developement of Ktulu!
Rahman Paidar:
Ktulu was originally developed in my spare time as a hobby work. I have still
the same interest to work on my program than before or even more. The fact that
one programmer can not make improvement does not mean he/she has lost his/her
interest. Success can be ahieved by experiment, collecting feedback, and you
know they all need time. Most of time, without getting failed, you can not reach
success and that's a fact.
Frank Quisinsky:
A half year ago you sent me different beta versions of Ktulu (Ktulu 5.2 /
6.0). Both versions are playing very interesting but it seems they are very
experimental. I believe you "search your style" for Ktulu?! I go further in the
question about Ktulu developement. I find out that you improved your engine in
different parts a lot. Unfortunately, the engine is weaker in other important
main parts and it's very difficult to see that your engine is in the final
result stronger. After all I have got the opinion, that you test a lot of different
ideas and you are not sure which way Ktulu have to go. Perhaps you got different
interesting ideas? Could you write a little bit to your Ktulu development
since the latest release version of Ktulu 5.1. The beta versions I have are too old and too experimental. I can't go in detail with my questions because it's not
possible for me, the beta versions are too "variable".
Rahman Paidar:
For Ktulu 6.1, I removed all changes since Ktulu 5.1 and added new code/idea/implementation
and etc.. There were lots of bugs and bad code programming which I didn't want
to see it in the program anymore. There are lots of changes and new code for
king safety, passed pawn evaluation, piece-development. The search function is
rewriten by new format of prunning. There are lots of new and original ideas in
this version. For years I have been working on some safe prunning system. It is
now become evidence for me chess programming theory needs some renovation. Some
concepts, are defined but could simply be misleading. For example Quiesence
search function is defined as a function which brings the position tactically to
a quiet position. Better is to defind it as a heavy prunning search function. I
don't want to go through details but one thing I can say you is that there are
lots of new ideas which would bring to your mind when you re-construct and think
about the basic concepts.
Frank Quisinsky:
Its not easy to
evaluate your engine. I believe this is the biggest compliment I can give you.
Your engine isn't easy to play with a lot of headstrong styles. I have to think
about
on Junior if I look in Ktulu games. Junior has an other style but it's also a
very headstrong engine. I am sure you are a strong chess player! Could you
explain a little your own interest on the playing style of Ktulu. Perhaps a
combination from strategy and tactic with many chess knowledge in middlegame?
Possible that I am totally wrong. Nobody can say more in detail to your engine as
yourself. For me Ktulu is the "Fidelity" under the young talents (Fidelity chess
computers by Kate and Dan Spracklen, USA). These chess computers have an own
style the customers like.
Rahman Paidar:
I am not a (ELO) rated player. Tactically I'm very weak, but I know very well
positional elements. Ktulu was initially designed to be a good tactical player.
I think most of chess programs are initially born to be a good tactical player.
It is easy for a programmer to say I have tactically improved my engine. But it
is most of time impossible or more harder to say I have positionally improved my
program. There are no strict defination for this stuff on chess theory. Only
Time can say you if you are wrong or right.
Frank Quisinsky:
Perhaps your main tester Jan Kiwitter will read my question and will
send me a short comment to Ktulu. Interesting are information about the playing
style and of course the opinion from your main tester Jan.
|
Jan Kiwitter (Germany):
Main tester of Ktulu!
German:
Ich habe den Eindruck, dass sich Ktulu verbessert hat (wie sehr, muss sich
zeigen - no more comment :-) ). Es hat jetzt mehr Schachwissen und spielt
(meistens) aktives, interessantes Schach. Typisch ist beispielsweise, auch mal
einen Königsangriff mit den eigenen Königsflügelbauern zu starten, was oft
klappt, aber auch ab und zu daneben geht, weil der eigene König dabei selbst zu
sehr geschwächt wird. Ktulu achtet jetzt mehr als früher auf schnelle
Figurenentwicklung und versucht, möglichst schnell offene/halboffene Linien mit
seinen Türmen zu besetzen. Wenn der Gegner passiv spielt, wird er in der Regel
schnell bestraft. Im Endspiel war Ktulu schon früher m. E. sehr erfolgreich, was
sich nicht geändert hat. Natürlich gibt es auch noch einige Schwächen: Manchmal
ist Ktulu zu optimistisch (was andererseits aber zum Teil für den interessanten
Spielstil verantwortlich ist), vereinzelt kommt es zu taktischen 'Black-outs',
wobei Ktulu aber andererseits auch taktische Schläge lange vor dem Gegner kommen
sieht, manchmal wird auch der angemessene positionelle Plan nicht gefunden (wie
bei jedem Schachprogramm ...). Ich hoffe, Deine Neugier wenigstens zum Teil
etwas befriedigt zu haben. Viel mehr kann natürlich Rahman selbst zu Ktulu
sagen!
English:
I have the impression that Ktulu has improved (how much is not clear so far - no
further comment :-) ). Now it has more chess knowledge and (most of the time)
plays active, interesting chess. Typical is for example to launch a kingside
attack with its own kingside pawns which often succeeds, but from time to time
also leads to a loss because it endangers the own king. The new version pays
more attention than before to a fast developpment of pieces and tries to place
rooks on open/half open files as soon as possible. If the opponent plays too
passive he will be punished quickly. Concerning the endgame in my opinion Ktulu
has been strong before and this has not changed. Of course there are also
weaknesses: Sometimes Ktulu is somewhat overoptimistic (on the other hand this
causes partly the interesting playing style), very rarely there are also
tactical black-outs, but otherwise Ktulu often sees tactical shots coming way
sooner than the opponent and finally from time to time Ktulu doesn't find a
suitable positional plan (which is the problem of every chess engine though).
German:
Ich gebe Dir noch 4 Beispielpartien der neuesten Beta (Zeitkontrolle jedes Mal
10min + 1sek; auf Laptop mit 1,9 GHz).
Viel Spaß damit!
|

Jan Kiwitter, Germany
Rahman's long time and main tester! |
Ktulu 6.1 Beta 3 - Ruffian 1.05
1.e4 e6 2.d4 d5 3.Nc3 dxe4 4.Nxe4 Nd7 5.Nf3 Ngf6 6.Nxf6+ Nxf6 7.Bg5 h6 8.Bh4 c5
9.Bb5+ Bd7 10.Bxd7+ Qxd7 11.Qe2 Be7 12.0-0-0 0-0 13.dxc5 Qc6 14.Ne5 Qxc5 15.Bxf6
Bxf6 16.Nd7 Qg5+ 17.Kb1 Rfd8 18.Rhf1 Qg6 19.g4 Rac8 20.f4 Rc6 21.h4 Rdc8 22.Rd2
Qh7 23.g5 Be7 24.Ne5 Rb6 25.Rd7 Bf8 26.Nxf7 Ba3 27.Nxh6+ Kh8 28.b3 Rbc6 29.f5
exf5 30.Nf7+ Kg8 31.Nd8 Qg6 32.h5 Qxg5 33.Nxc6 Rxc6 34.Qe8+ Kh7 35.Rd8 Qxd8 1-0
Aristarch 4.41 - Ktulu 6.1 Beta 3
1.c4 Nf6 2.Nc3 e6 3.Nf3 Bb4 4.Qc2 0-0 5.a3 Bxc3 6.Qxc3 b6 7.g3 Bb7 8.b4 d5 9.Bb2
dxc4 10.Qxc4 Bxf3 11.exf3 Qd5 12.Bxf6 Qxf3 13.Bxg7 Kxg7 14.Rg1 Nd7 15.Bg2 Qf6
16.d4 Rad8 17.Ke2 Ne5 18.Qxc7 Nd3 19.Kxd3 Qxd4+ 20.Ke2 Rc8 21.Qc1 Qe5+ 22.Qe3
Rc2+ 23.Kf3 Rc3 24.Rac1 Rxe3+ 25.fxe3 Qf6+ 26.Ke2 Qb2+ 27.Kf3 Rd8 28.Rgd1 Rxd1
29.Rxd1 Qxa3 30.Rd4 Qc1 31.Rg4+ Kf6 32.Rf4+ Ke7 33.Rh4 Qd1+ 34.Kf2 Qc2+ 35.Kf3
a5 36.bxa5 bxa5 37.Bf1 a4 38.Be2 a3 39.Bc4 Kd6 40.Rd4+ Kc5 41.Bd3 Qd1+ 0-1
Ktulu 6.1 Beta 3 - Pharaon 3.2
1.d4 d5 2.c4 c6 3.Nf3 Nf6 4.Nc3 e6 5.e3 Nbd7 6.Bd3 dxc4 7.Bxc4 Nd5 8.0-0 Be7
9.e4 Nxc3 10.bxc3 Qa5 11.Bd2 e5 12.Qb3 0-0 13.a4 Bf6 14.Rfb1 Qb6 15.Qc2 Qc7
16.a5 Be7 17.Rb2 Bd6 18.Be3 Nf6 19.dxe5 Bxe5 20.Nxe5 Qxe5 21.f3 Qc7 22.Qb3 Rd8
23.Bd4 Ne8 24.Rd2 Rd7 25.Rdd1 Rd6 26.Rab1 Rh6 27.e5 c5 28.Be3 Rc6 29.Bf4 Qe7
30.Bd5 Ra6 31.Qa4 Kh8 32.Bg5 Qf8 33.Bc4 Rc6 34.Bb5 Re6 35.Rd8 Rxe5 36.Bf4 Re6
37.Bxe8 Bd7 38.Bxd7 Qxd8 39.Bxe6 fxe6 40.Rxb7 Qd3 41.Rxa7 Qb1+ 42.Kf2 Rg8 43.Be5
Qc1 44.Rd7 Qb2+ 1-0
List 5.12 - Ktulu 6.1 Beta 3
1.Nf3 Nf6 2.g3 d5 3.Bg2 c6 4.0-0 Bg4 5.d4 e6 6.Ne5 Bf5 7.f3 Bd6 8.Nc3 c5 9.e4
cxd4 10.Qxd4 Bxe5 11.Qxe5 Nc6 12.Qf4 Bg6 13.exd5 Nxd5 14.Nxd5 Qxd5 15.c3 0-0
16.Be3 Qb5 17.Rf2 Rfd8 18.a4 Qb3 19.Bf1 Bd3 20.Bg2 f6 21.Bh3 Bc4 22.Qh4 b6
23.Qe4 Rd6 24.Rd2 Rxd2 25.Bxd2 Rc8 26.Bf4 g5 27.Bd6 Bd5 28.Qe2 Rd8 29.Ba3 Ne5
30.f4 Nf3+ 31.Kf2 f5 32.Bg2 Nxh2 33.Qh5 Ng4+ 34.Kg1 Kh8 35.Qxg5 Rg8 36.Bxd5 Qxd5
37.Qh4 Qd2 38.Be7 Rg6 39.Bg5 Qxb2 40.Rf1 Qxc3 41.Qh3 Qe3+ 42.Kh1 Qe4+ 43.Qg2
Qxg2+ 44.Kxg2 Ne3+ 45.Kf2 Nxf1 46.Kxf1 Rxg5 47.fxg5 Kg7 48.Ke2 Kg6 49.Kf2 0-1
Frank Quisinsky:
Since version 5.0 your
engine is commercial available. Are you disappointed about the
press? It seems the press only likes to power ChessBase products. Today young
talents get a very bad start by the press and of course from possible
customers. Customers await a free engine and the most don't respect the way the
author go with commercial ideas. In my opinion it's a pity that customers
download free engines without to look in the engine and buyed boring chess
programs in version 6, 7, 8 which are not stronger than the preview versions. To
trying to make a little bit for computer chess area is an offsite route. Today
different groups don't work hand in hand and the complete area
is unorganised. In these times engine programmers have bigger problems to make
their program commercial available. The expectations of customers are too high and
most of them are not able to find a pearl. So we got with Ktulu and your person two
interesting parts in this area. A program which goes an own but very interesting
way (speaking from the playing style of Ktulu) and a programmer with a realistic opinion in an
unrealistic time. What is your opinion about the situation? Do you know that a
lot of persons are working with a cracked version of Ktulu? What is your comment
to such illegal piracy?
Rahman Paidar:
First let me say the world of computer
chess and chess programs including chess products, programmers and other people,
are exteremly different than other areas. We have so many strong free programs
in front of us and it makes it completely difficult for a person to make his/her
program commercial. You know writing a chess program is not easy. There are less
than perhaps 1000 persons that knows how to do it. Chess programmers need
feedback and they also want to exchange some information between themselves. And
it is the point in my opinion that let the programmer to improve slowly their
programs by making it freely available. I know some people are working on the
cracked version of Ktulu. It is not of course important for me. We had the same
situation only one week after the first release (Ktulu 5.0). Unfortunately there
are always such problems everywhere at anytime. I think we have enough trusted
customers who has buyed Ktulu, and it is more important for me.
Frank Quisinsky:
In my latest questions
you could find a lot of negative sentences. This is only one point of view. The
other and clear positive point of view is your engine. Again, you wrote me that
you will offer Ktulu 6.2 soon. This isn't an update for Ktulu 5.1
customers. I have no information about your latest version of Ktulu and perhaps
you can write a little bit about the new ideas you added in Ktulu 6.x. How much
stronger is your new version in your own tests? What is the most interesting
point in chess programming in which you spent the most of your time in developing since
the latest available version 5.1? It's very interesting to know more about your engine and the main ideas you
have. Could you give us information about the changes in program structur
since Ktulu 5.1 too?
Rahman Paidar:
Well, as I said already, the changes from Ktulu 5.1 to Ktulu 6.2 is so much to
name some of them, King safety and passed pawn evaluation are completely
rewritten. Most of dangerous attacks are seen in evaluation function. This makes
Ktulu stronger both in tactics and positional playing. I have removed some
unsafe prunning algorithms and replaced them with better methods. There are so
many works done on the search function and Qsearch functions. During this time I
found so many original ideas and you can clearly see how Ktulu is tactically
improved.
|
 |
Frank Quisinsky:
In the latest months a
lot of strong and free engines are clearly improved or completely new. Speaking
from SlowChess, Spike, Glaurung, Fruit and many others. Last weekend I spoke
with a friend about all the available developments. He gave me the information
that he have more fun with computer chess in the time RexChess (on of the first
available chess programs for MS DOS) are ready for take off. My friend played
many games with RexChess. He gave me the information that he missed the latest
energy in computer chess today. Press reviews are boring and allways the same and
also
computer chess fora are today not very interesting. Today we have an oversupply
is his opinion. We can find information about eng-eng results but only in "good days" information in detail about the engines and programmers. I will
give his question to you Rahman. Do you believe, that it's more
interesting to busy oneself with certain programs or with many of the
available programs? To know one program in detail can be more interesting as playing with 50 engines tournaments!
How do you test Ktulu? Eng-Eng
games against the others, test suits or do you search in Ktulu games by tester
mistakes? Which tip of testing can you give the groups of users?
Rahman Paidar:
Well, taking into account all possible chess programs and their version numbers
and so on, we may completely run out of time if we want to participate all of
them :) Of course we can devide them into several sections and parts and make
some tests. I believe each chess program has an own style of playing. You may
have observed that one program for example chooses Ne5 in one position but in
other hand, the opponent was expecting Fg5 and these two moves are interesting
and playable. To make short story, I think every programs has merit to be payed
attention like each individual person. For me it is hard to include several
engines and test a new version of Ktulu against them. Most of time I choose one
or two strong opponent and test Ktulu against them. Of course we then focus more
on the games and play so many games. Unfortunately I can not give all details
about how I test Ktulu, there are so many games which would be played and of
course the games are not started at opening nor end at checkmate. Depended on
which changes I have done in the program, the test may vary, for example if I
have changed passed pawn evaluation, then the phase game will start at
middlegame and will finish at checkmate but if I have changed the kingsafety
evaluation, the phase game will start at middle game and will finsih when one
side get ahead in material or get a winning position. Sorry I can not give more
details :)
|
Frank Quisinsky:
In your country it's
for sure not easy to find other persons with same interest. I think computer
chess in Iran isn't much popular. Do you have friends for an "information
exchange" in your country. This could helps you to find new energy and motivation!
What are your friends saying about your hobby "computer chess"?
Rahman Paidar:
No, I have not chess friend in Iran at the moment. Only, my brother can play
decent chess but unfortunately he is not in my town. He is studying in another
town at university. I played chess when I was 7 and since then I read so many
books about chess. When I was younger (about 8 years ago) , I went to chess club
at weekend and had so many friends, but now I have lost them all. You know what
your friends tell you when your hobby is different than theirs, "you are mad", "you
are wasting your time", ... :-)
Frank Quisinsky:
Again, how many stronger is Ktulu 7.0 :-)
Rahman Paidar:
Unfortunately I can not measure the ELO difference between two version. If I
want to make my opinion I would say, more than 20 ELO, is a good safe margin.
Better is to wait what says other testers and customers.
Frank Quisinsky:
Arena programmer
Martin Blume is working all the time on Arena Chess GUI. So far he is very busy
with the ideas of our Gladiators. Linux support, 3d options, database options
for some examples. He will look
and find out what is interesting for the future of Arena Chess GUI. The first alpha I
have got is very interesting. You know that I have the idea to
produce an Arena CD for the mass market. It's a little bit dangerous to do this with
a free available product but the ideas we have are too interesting. The main engines
for such a project are SOS for Arena, Ktulu and Patriot. Your engine is the
prime example for an exceptional offer. But the project needs time and I will wait
for Arena 2.0. Important is the compatibitliy of Ktulu. Fischer Random Chess
(Chess960) or UCI II support with the new six available possibilities awaiting
of customers.
Do you gave UCI-II the first look? Are you interesting in Chess960?
Rahman Paidar:
I never heard anything about Chess960, sorry. Supporting UCI II is easy when you
have implemented UCI support already. The only problem is time, and priority.
For example when you want to improve your program and have a very limited time,
then your first prior to work is improving chess engine, adding new knowledge
and etc, and UCI II support will be postponed to a proper time.
Hint:
Information to Chess960 can be found on the
webpage by
Reinhard
Scharnagl
Frank Quisinsky:
Rahman, what are your
other main interest. Do you have other additional hobbys or is there only computer chess? The
programmer of Naum, Aleksandar Naumov, asked me to try to find out more
about engine programmers. I will try it and perhaps you can write something
about:
Age, birth place, profession, education ... or other things about your persons. Each
information is great for us.
Rahman Paidar:
Electronics and Aritificial Neural
Networks (INN), are my main interest excluding chess programming. I was born on
April 1982 in Tehran, and I am still studying at university (Electrical
Electronics).
Frank Quisinsky:
Thanks for your answers
and good luck for your Ktulu. If I will get more time I would like to do more for
Ktulu but here are three childrens. My girlfriend accepts my hobby but in
German we say:
"We have to hold the church in the
village". I hope that many users will support the young commercial programmers,
like Gian-Carlo Pascutto (Deep Sjeng), Vladimir Yelin (Patriot) and others which
choose the commercial way. Today Ktulu is a part of our area and have a good name.
Rahman Paidar:
Well, to be honest, your questions to answer were as hard as an university's
exam for me :-)
Thank you too for your time and keep up your good work.
Go to
News-Ticker: Main Page,
Page 1,
Page 2,
Page 3,
Page 4,
Page 5, Page 6,
Page 7,
Page 8,
Page 9 |