[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Results of 1998 Fost cup
The results are attached. I'll have an updated ranking chart
soon.
Enjoy,
David
The 1998 FOST Cup Open Computer Go Championship
The FOST cup was held at the Nihon Ki-in in Tokyo, Japan, on August
28 and 29. The format was a little different this year, with
three separate competitions. On the first day, a 6 round swiss
tournament selected 4 programs to advance to the final tournament.
On the second day, the top four programs from last year and the
four qualifiers had a 7 round, round-robin tournament. The top
four programs from last year (Handtalk, Go Intellect, Go4++ and Silver
Igo) did not participate in the first day qualifying
tournament.
On the second day there was also an optional consolation tournament
for programs that did not qualify for the final 8. There were 41
programs registered, from 9 countries, with 13 new ones that had not
participated in FOST before. There was a moderate earthquake early
on the second day, but no games were affected.
There were a lot of changes in the order of finish from last year. In
general, programs written by teams improved greatly (Silver Igo, Hamlet,
Wulu, Young Leaf). The top two programs of last year both dropped,
after both had major changes made during the year.
Prize money was (in Yen):
1) 1,000,000
2) 400,000
3) 300,000
4) 250,000
5) 200,000
6) 150,000
7) 100,000
8) 80,000
The first day qualifiers in order of finish, were:
1) Jimmy
2) Wulu
3) Many Faces of Go
4) Hamlet
The final result was:
1) Silver Igo Silver Star Lab D.P.R. Korea
2) Hamlet (Hit) Ji Il Kim
Alexei Telitsine
Cheol-Jin Park Korea
3) Goemate (Handtalk) Chen Zhixing China
4) Go4++ Michael Reiss England
5) Wulu Lei Xiuyu
Chen Guobao
Lu Jinqiang
Li Zhihua China
6) Many Faces of Go David Fotland USA
7) Jimmy Shi-Jim Yan Taiwan
8) Go Intellect Ken Chen USA
Silver Igo received a 3 Kyu rank certificate and trophy.
>From the qualifying tournament:
9) Young Leaf (Wakaba) Junji Yatsuzuka/Tei Meiko Japan
10) Goro Yasuo Oishi Japan
11) Igo Jiman 1.0 Keiji Takashima Japan
12) Go Master Jee Wonho Korea
13) Gogol Tristan Cazenave France
14) Monkey Jump Hiroto Yoshii Japan
15) Katsunari Shinichi Sei Japan
16) Fun Go Yong-Goo Park Korea
17) K1 Lin Gwan-Jae Taiwan
18) Igo 98 Noriaki Sanechika Japan
19) Aya Hiroshi Yamashita Japan
20) Martha Ichiro Ujiie Japan
21) Baduk-1 Lim Jaebum Korea
22) Explorer Martin Mueller Austria
23) Gizumogo Hirofumi Iwasaki Japan
24) B/W Osamu Ushio Japan
25) Test2 Bunsen Sen China
26) Heian Igo Yoshitaka Koiyama Japan
27) Deer Stallion 2 Atsushi Yoshika Japan
28) Takuchan Takuo Tabuchi Japan
29) Utoro Tuneo Horii Japan
30) Darren's program Darren Cook England
31) Mutsuki Takahisa Yoshida Japan
32) Tokyo 98 Toshikazu Sato Japan
33) Gosaku Tamizo Shiraishi Japan
34) Twigo Tetsuya Wakamatsu Japan
35) Deer Stallion 1 Takuya Kojima Japan
36) Keeping Awkaward I Yung Jye Huang Taiwan
37) Go San Jin Kuya Fukuzawa Japan
38) Caren's Whisper Katsumi Kobayashi Japan
The time limits were 50 minutes per program before the game reached
move 250. To fit 6 rounds in one day at that time limit, the
preliminary tournament used accelerated pairings. The pairings for
the next round were completed before the games of the previous round
were finished. A pure swiss system would have had slightly different
pairings. Caren's Whisper dropped out after two rounds, so there were
an odd number of programs for the last four rounds.
The rules give the bye to the strongest programs from last year, so in
the last round, when Jimmy and Hamlet were the only undefeated programs,
Jimmy got a bye to remain undefeated, and Hamlet was paired with Many Faces.
When Many Faces beat Hamlet, it ended in a tie for 4th place with Young Leaf.
If one of Young Leaf's weaker opponents had one more win, Hamlet would not
have qualified. There was a playoff game between Hamlet and
Young leaf, which Hamlet won.
With a large number of programs and few rounds, there is no guarantee
that the top four programs will actually be selected for advancement
to the second day tournament. But this year several people told me that
they felt the first day tournament had chosen the right programs.
The consolation tournament was single elimination, rather than swiss,
so perhaps its rank ordering is not as accurate as the main tournament.
The results were:
1) Fun Go
2) Goro
3) Katsunari
4) Young Leaf
5) Go Master
6) Gogol
7) Monkey Jump
8) Aya
9) Explorer
10) Mutsuki
11) Twigo
12) K1
13) Test2
14) Utoro
15) Deer Stallion 1
16) Deer Stalion 2
The final tournament grid (in initial seed order):
opponent: 1 2 3 4 5 6 7 8 Wins SDS Place
1) Goemate (Handtalk) X + - - + + + + 5 13 3
2) Go Intellect - X + - - - - - 1 8
3) Go4++ + - X - + + - + 4 4
4) Silver Igo + + + X + + + - 6 1
5) Jimmy - + - - X - + - 2 3 7
6) Wulu - + - - + X + - 3 5
7) Many Faces - + + - - - X - 2 5 6
8) Hamlet - + - + + + + X 5 14 2
The top programs were closer in strength this year than before. Go
Intellect, (finishing 8th) beat Go4++ (finishing 4th), which beat
Goemate/Handtalk (finishing 3rd), which beat Hamlet (finishing 2nd),
which beat Silver Igo (1st).
Chen Zhixing spent 5 months last year rewriting Handtalk with a
new pattern management system to enable much higher program strength
and more patterns. He estimates 3 years to complete conversion to
the new system. So this year's Handtalk is not much stronger than
last year. It's for sale in the USA as Handtalk, and in Japan as
Shudantaikyoku 2.
Go Intellect also has a major change this year. Evaluation is now based
on the temperature of local fights, looking at each position with
both players moving first. Before the tournament Ken estimated that the
new program was stronger than the old, but that it had some severe
tactical analysis bugs that could cause problems. The tactics bug
showed up in every game except against Go4++ (since Go4++ doesn't fight
much) and Go Intellect won only against Go4++. It's for sale in the USA
as Goddess, in Korea as Kuksu, and in Japan as Heiei_kiin, Koutekisyu.
Go4++ is still focussed mainly on accurate territory evaluation and
is the weakest at tactics. Michael tunes extensively against Handtalk,
playing about 5000 test games last year, and reaching an 83% winning
percentage. It payed off for him with a win against Handtalk by 1.5
points in a quiet game, but he threw away an easily won game against
Many Faces due to tactical blunders in the endgame. It's for sale
in Japan as "The Strongest Game of Go".
Silver Igo is stronger than last year, and has a team of people working
on it full time. But it still lost to Hamlet and won a close game
against Many Faces.
Since Jimmy placed well last year, it got easier pairings in the
preliminary tournament, and finished undefeated, with 6 wins. It didn't
play any of the 4 programs that finished with 5 wins. In the final,
it beat Many Faces by 2.5 points.
Wulu is being developed by a team of 4 programmers that includes one
of professional strength. This was Wulu's first appearance at FOST,
but it competed in the 1996 Ing tournament and finished 8th.
Many Faces is still strong comparatively at life and death. It killed
groups in many games. But it is not good at keeping territory. Still,
it beat Hamlet once, and had a close game with Silver Igo, and lost to
Jimmy by only 2.5 points. I took a break from computer go last year,
and didn't work on Many Faces between November and April, then I took
another 6 weeks off to develop Igowin. So Many Faces was not as strong
as it could have been. It's for sale in the USA, and in Japan as AI Igo 6.
Hamlet has a team of 3 programmers and 2 to 4 go experts. It's stronger
than last year, when it was called Hit, but lost to Many Faces in the
preliminary tournament and beat Wulu by 2.5 points. It was the only
program to beat Silver Igo. It's on sale in Korea as Medusa, and has sold
about 8,000 copies to date.
The most interesting program is Young Leaf. It's written by a small
team that includes a professional 9 dan. It was by far the fastest
program I played. It made all of its moves in only 90 seconds, but it
plays a very good game. I haven't seen a program that can play so well
with so little reading since the early versions of Handtalk. When
they add better tactics to it, it could be very strong. Young Leaf
only lost to Many Faces and Hamlet (in a 7th round playoff).
Aya uses an eye-pattern database, using the ideas developed by Dave
Dyer, with 1,289,007 eye patterns.
K1 and Igo Jima both did very well for their first year.
Many of the programs that finished well last year dropped a lot this
year, but Goro held up and finished 6th.
Go Master improved a lot since last year.
Keeping Awkaward I played reasonable game, but lost every game by
running out of time.
The Main tournament grid (initial order based on results from last year):
round 1 2 3 4 5 6 wins place
1) Gogol -19 +11 +32 +21 -15 +13 4 9
2) Aya -20 +10 +30 +32 +21 -8 4 15
3) Fun go +21 +13 -25 -28 +32 +7 4 12
4) Jimmy +22 +12 +14 +8 +31 +32 6 1
5) Many Faces +23 +15 +27 -25 +8 +6 5 3
6) Hamlet +24 +14 +16 +19 +25 -5 5 4 (after playoff)
7) Explorer -25 +17 +21 -15 +18 -3 3 18
8) Goro +26 +16 +20 -4 -5 +2 4 6
9) Gosaku -27 -18 -17 +30 +36 -29 2 29
10) Mutsuki -28 -2 -26 +36 +30 -11 2 27
11) Igo 98 -29 -1 +23 +17 +20 +10 4 14
12) Katsunari +30 -4 -28 +22 +29 +37 4 11
13) B/W -31 -3 +33 +23 +17 -1 3 20
14) Gizumogo +32 -6 -4 +26 -16 +22 3 19
15) Young Leaf +33 -5 +29 +7 +1 +31 5 5 (after playoff)
16) Monkey Jump +34 -8 -6 +37 +14 +27 4 10
17) Darren's program -37 -7 +9 -11 -13 +20 2 26
18) Tokyo 98 +36 +9 -31 -27 -7 -33 2 28
19) Go Master +1 +36 -37 -6 +27 +26 4 8
20) Twigo +2 -28 -8 -31 -11 -17 1 30
21) Takuchan -3 +29 -17 -1 -2 +23 2 24
22) Utoro -4 +30 +36 -12 -26 -14 2 25
23) Keeping Awkaward I -5 -31 -11 -13 -33 -21 0 32
24) Raisin -6 (crashed, then withdrew)
25) Wulu +7 +33 +3 +5 -6 +28 5 2
26) Heian Igo -8 +32 +10 -14 +22 -19 3 22
27) Baduk-1 +9 +37 -5 +18 -19 -16 3 17
28) Igo Jiman 1.0 +10 +20 +12 +3 -37 -25 4 7
29) Deer Stallion 2 +11 -21 -15 -33 -12 +9 2 23
30) Deer Stallion 1 -12 -22 -2 -9 -10 +36 1 31
31) K1 +13 +23 +18 +20 -4 -15 4 13
32) Caren's Whisper -14 -26 -1 (dropped out) 0 34
33) test2 -15 -25 -13 +29 +23 +18 3 21
34) -16 (no show)
35) (no show)
36) Go San Jin -18 -14 -22 -10 -9 -30 0 33
37) Martha +17 -27 +19 -16 +28 -12 3 16
-David Fotland