[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: 5th FOST Cup results and commentary
Hi David,
Thanks for the report and commentary.
It's kind of exciting to see that so many programs are ahead of
HandTalk. HandTalk is a very impressive program, and it's one of the few
go programs I have tried. I am also glad to see Many Faces to be on the
top three after fierce competition.
Could we, especially those who took part in recent competitions,
chat about the new trends/techniques in go programming, why the go
programs are getting so much better (or not), why rewriting your program
easily hurts you, ...
Is it likely that, since HandTalk has become a target for all
the programmers, it didn't "perform well"?
Thanks in advance.
-- Mousheng Xu
-----Original Message-----
From: David Fotland [mailto:fotland@xxxxxxxxxxxxxxxxx]
Sent: Sunday, September 05, 1999 6:51 PM
To: computer-go@xxxxxxxxxxxxxxxxx
Subject: 5th FOST Cup results and commentary
The 1999 FOST Cup World Open Computer Go Championship
The FOST cup was held at the Nihon Ki-in in Tokyo, Japan, on September
3 and 4. The format was very different this year. The first four
FOST cups were open tournaments, attracting about 40 programs.
This year they decided to limit the final tournament to 16 programs.
The top 8 finishers from last year were seeded, and there was
a preliminary tournament in July to select the other 8. The
final tournament is an 8 round swiss, played over two days, with
accelerated pairings.
The tournament rules had some changes as well. We arrived in
Japan to find that FOST had decided not to allow use of the
serial protocol. Instead, the operators had to enter moves
by hand, with a 3rd person at each game running a clock. If the
operator made an error entering a move and didn't detect it before
entering his/her next move, that program loses. Time limits
are 50 minutes for 125 moves each. There was only one game
lost due to a data entry error, by Haruka against KCC Igo.
No games were lost on time, and only one game was adjudicated
by the Professional, when 50 minutes expired after move 125.
The competition went very smoothly. NEC provided very small,
slim tower computers that were very easy to move between rounds.
Ken Chen did another major rewrite of Go Intellect last year and
changed the name of the program to Go Intellect II.
Kim HyunSoo works on FunGo full time, but the program is not
commercialized. It has become very strong in the last year.
It only lost to Wulu by 0.5 points, it was the only program
to beat KCC Igo, and it was ahead most of the game against
Many Faces. Since the author is working on it full time, I
expect it will have a good chance to take first place next year.
Go4++ and Go Intellect II used 600 Mhz Pentium II machines.
Most other programs ran on the 333 Mhz NEC machines provided by
FOST and NEC.
Goemate was not any stronger than the old Handtalk, so
Prof Chen used the older Handtalk engine. In tests before
the tournament, the old Handtalk beat the new Goemate
8 out of 12 games. Handtalk won first place in seven
consecutive tournaments, but could only finish seventh
this time. This demonatrates how quickly computer go is
progressing.
Explorer did not participate because Martin Mueller was not
informed of the qualification tournament in time.
KCC Igo is written by the KCC Igo team in North Korea. This team
includes people who worked on Silver Igo, and uses the Silver Igo
interface, but we were assured that there was no plagiarism involved.
Last year Silver Igo was accused of plagiarizing Handtalk, but
FOST did not make a determination, since they did not have the
executable. Prof. Chen later got a copy of the commercial
version of Silver Igo, and believed it to be plagiarized. This year
there was an accusation of plagiarism against this program by
Michael Reiss. FOST required all participants to give
a copy of their executable in case there was an accusation of
plagiarism, so they should be able to resolve this plagiarism issue.
10 people at KCC work on strategy games, chess, shogi, etc.
They have one full time go programmer, Jong Song Hwa, who has
been working on this engine for 3 years. They test against AI Igo,
which has the Many Faces of Go engine, and other commercial
Japanese programs. It plays a similar style to last year's Silver
Igo, and seems much stronger than the program of the same name
that was entered in the CGF computer go championship. But the
KCC team traded programs with several others after the championship,
and vigorously denied that the new program has any relationship to
Silver Igo when I asked them (through two interpreters, English
to Japanese to Korean).
There were several new programs:
Masayan, by Masatsune Koshida, from Japan.
Don, by Syoji Hiasa, from Japan.
Haruka, by Ryuichi Kawa, from Japan. This program came in
second place in a world computer go contest in July, and is
expected to do very well. The author has been working on
the program about 5 years full time, but didn't enter this
competition until he thought he had a chance to win. It has
been commercialized for a couple of years.
Prizes were (in Yen):
1) 1,000,000, certificate, trophy, plaque, and vase.
2) 400,000, certificate
3) 300,000, certificate
4) 250,000
5) 200,000
6) 150,000
7) 100,000
8) 80,000
The final result was:
1) KCC Igo Silver Star Lab D.P.R. Korea
2) Go4++ Michael Reiss England
3) Many Faces of Go David Fotland USA
4) Haruka Ryuichi Kawa Japan
5) FunGo Kim Hyun Soo Korea
6) Wulu Lei Xiuyu
Chen Guobao
Lu Jinqiang
Li Zhihua China
7) Goemate (Handtalk) Chen Zhixing China
8) Go Intellect II Ken Chen USA
9) Go Master Jee Wonho Korea
10) Goro Yashuo Oishi Japan
11) Wakaba Tei Meikou Japan
12) Jimmy Shi-Jim Yan Taiwan
13) Masayan Masatsune Koshida Japan
14) Biwako Masahiro Tanaka Japan
15) Don Syoji Hiasa Japan
16) Mutsuki Yoshida Takahisa Japan
KCC Igo was given a 2 Kyu diploma by the Nihon Kiin, signed by Otake.
The top seven programs seem to be very similar strength. Many of their
games were very close, and the order among these programs was very
different
in the CGF tournament in July:
>From CGF:
1) Go4++
2) Haruka
3) Goemate
4) Fun GO
5) Wulu
6) Many Faces
7) KCC Igo
Go4++ has more consistent results, since it plays for territory and
avoids
taking any risks by fighting, so perhaps it is slightly stronger than
the
rest.
The tournament grid, in seed order.
round: 1 2 3 4 5 6 7 8 Win SOS Place
1) KCC Igo +9 +5 +6 +3 +2 +10 -12 +16 7 38 1
2) Goemate +10 +6 -5 +4 -1 +9 -16 -3 4 41 7
3) Go4++ +11 +7 +16 -1 +4 +5 +9 +2 7 37 2
4) Wulu +12 +8 -7 -2 -3 +13 +11 +10 5 32 6
5) Many Faces +13 -1 +2 +16 +9 -3 +10 +12 6 38 3
6) Jimmy +14 -2 -1 -9 -10 -16 +15 +13 3 28 12
7) Go Intellect II +15 -3 +4 -10 -16 -12 +8 +11 4 33 8
8) Wakaba -16 -4 -12 +11 +13 +15 -7 +14 4 25 11
9) Goro -1 +13 +14 +6 -5 -2 -3 +15 4 31 10
10) Go Master -2 +14 +15 +7 +6 -1 -5 -4 4 31 9
11) Masayan -3 +15 -13 -8 -12 +14 -4 -7 2 29 13
12) FunGo -4 -16 +8 +14 +11 +7 +1 -5 5 34 5
13) Biwako -5 -9 +11 +15 -8 -4 -14 -6 2 26 14
14) Don -6 -10 -9 -12 -15 -11 +13 -8 1 25 15
15) Mutsuki -7 -11 -10 -13 +14 -8 -6 -9 1 24 16
16) Haruka +8 +12 -3 -5 +7 +6 +2 -1 5 40 4
All game records will be available from the FOST web site.
http://203.183.116.3/FOST/02GO_E.HTM
Since there was no cable connection, I was busy during each
round operating Many Faces, but here are some of the interesting
details I noticed:
Round 1:
Many Faces beat Biwako by 90.5
Wulu beat Fungo by only 0.5 point
G04++ beat Masayan by 103
Goemate beat Go Master by 2.5
Round2:
Ken Chen thought Go4++ couldn't fight well, so he played Go
Intellect at a low level to avoid time problems. But Go4++
is much better at fighting now, and it killed a group and
won.
Round 3:
Many Faces beat Goemate after a big fight where both
sides made big mistakes.
The Go4++ and Haruka game went back and forth, but Go4++ won after
killing a big group.
Go intellect beat Wulu.
Round 4:
Many Faces scored a big upset against Haruka. Haruka was way ahead,
and Many Faces almost ran out of time, but Haruka let Many Faces
start a ko for the life of a big group, and lost it. Many Faces
played the last 100 moves at level 3, in less than 2 minutes,
including the game-winning ko.
Go intellect lost to Go Master by 3.5 in time trouble.
Round 5:
KCC Igo beat Goemate by 5.5. Go4++ beat Wulu by 8.5
The game between Haruka and Go Intellect turned into a huge
semeai, worth over 100 points. Go Intellect had several chances
to win it, but in the end Haruka had a big win.
Round 6:
Go4++ beat Many Faces by 7.5. Go4++ sacrificed two large groups
in the opening to get a big center. Many Faces was quite a bit
ahead, but failed to reduce the center effectively.
FunGo and Go Intellect both use almost all their time, so they
were the last game finished. It was very close, with Go
Intellect leading by about 1 or 2 points until the late endgame.
But Go Intellect started throwing stones into enemy territory
and gave up points, losing by 9.5.
The game between Don and Masayan had to be adjudicated by the
pro since Don used all its time after move 250.
Round 7
The game between FunGo and KCC Igo was close going into the endgame,
but KCC captured some stones and was ahead by over 50, then
FunGo captured something and won by the komi. This upset meant that
if KCC lost to Haruka in the last round, there would be several
programs tied for first place, so tiebreakers would determine
the winner.
Many Faces beat GoMaster by 99.5, killing 4 big groups.
Round 8:
Haruka made a move entry mistake and didn't catch it for several
moves, so he lost to KCC Igo, giving KCC Igo an uncontested win.
Many Faces was behind against FunGo, but caught up in the endgame
and won by 7.5.
After the tournament, Kojima Takaho 9 Dan, gave commentary on the
games between Many Faces and Haruka, and beween Go4++ and KCC Igo.
He said computers are getting close to Dan level, at about 4 kyu.
-David Fotland