US Go Congress 1999

Date: August 1-2, 1999

Place: San Francisco, USA

Operator: Daniel Bump

Version: GNU Go 2.3.73

Results:

1. Many Faces of Go
2. GNU Go
3. Smart Go
4. Team Go

Excerpts from the mailing list:

Date: Sun, 1 Aug 1999 21:06:58 -0700
From: Daniel Bump
To: gnugo@match.stanford.edu
Subject: GNU Go at the US Go Congress

[...]
After lunch was the computer tournament. Only 4 programs
entered, and Many Faces defeated us. Unfortunately Anders
Kierulf had a conflict and we'll have to play with his
program (Smart Go) tomorrow. I'll go in to San Francisco
twice, in the morning to play in the US Open, then in
the evening to play with Anders' program.

The real plus for me was long interesting converations
with David Fotland and Mark Boon. I bought a fresh copy
of Many Faces (and a copy of Handtalk) and David says he'll
give me a binary of the latest version.

I won't post the games right now because they're on the
laptop, but they will make good tuning material. Many Faces
is stronger, and Team Go (the other opponent) is extremely
weak, but just as with the game against Poka, a game against
a weak opponent can be good for pointing out holes in the
pattern database.

Date: Mon, 2 Aug 1999 06:11:18 -0700
From: Daniel Bump
To: gnugo@match.stanford.edu
Subject: more about the Congress

A few things I forgot to mention in my last e-mail.

The null modem cable worked. We were able to connect
my Linux laptop with David Fotland's Windows one with
a null modem cable and play a game automatically.

The Computer Go Federation tournament was held
recently in Japan. As you may recall, we thought about
participating but unfortunately the proposed "open source"
section didn't materialize, so we stayed home.

Handtalk placed third. Go++ (recent version) placed
first, and second was a new Japanese program.

Last night before I went to sleep I booted up MSDOS and
played a 9 stone game with Handtalk. It uses free
placement for the handicap stones, made some shimaris
and extensions for its handicap moves.

I found that I was able to kill stuff easily but on the
other hand it consolidated a moyo and was able to kill
my invading stones, and ended up winning. The life and
death code seems to be somewhat better than ours, but
not that much better. On the whole it is a lot stronger
than GNU Go, and a lot stronger than Many Faces.
The executable is small, only about 120K as I recall.

It will be interesting to get a few games with Handtalk
now for tuning.

Date: Mon, 2 Aug 1999 17:35:01 -0700
From: Daniel Bump
To: gnugo@match.stanford.edu
Subject: US Go Championship games

The US Computer Go Championship had only 3 entries this year,
Many Faces of Go (David Fotland), Smart Go (Anders Kierulf)
and Team Go (Gary Boos). Team Go is very weak.

We lost to Many Faces, but won against Smart Go and Team Go.
So I suppose we will have second place.

All three games are available by anonymous ftp to match.stanford.edu,
in pub/ the games are called:

usgo-mfgo.sgf
usgo-smart.sgf
usgo-team.sgf

I recommend the Smart Go game. We won a lost game because
the reading code worked and the life and death code worked,
so when an opportunity came up we were able to take advantage
of it. The game seemed won by Smart Go until ...

There is plenty of rich tuning material in these games.
It is instructive to see how we were outplayed by Many Faces,
and also by Smart Go (up to the point where Smart Go blew it).
Our program needs LOTS of tuning, far more than it needs
new code. We are now OK at making moyos, but we are not
good at reducing them. In the game against Smart Go, we
invaded his moyo, far too late, nearly lived, but died
because of too many tenukis.

Anders Kierulf, the author of Smart Go, works for Microsoft
and lives here in the area, but he says he is going to quit
his job to work on the Go program full time. He, David Fotland,
and Mark Boon were all very interesting to talk to about Go
programmming. Mark Boon no longer works on Goliath, but he
is very smart and knowledgeable.

Dan

Date: Sun, 8 Aug 1999 08:48:55 -0700
From: Daniel Bump
To: gnugo@match.stanford.edu
Subject: GNU Go at the US Go Congress

GNU Go won two prizes at the US Go Congress: SECOND PLACE
(behind Many Faces of Go) and BEST NEW PROGRAM. The awards
ceremony was last night at San Francisco State University.

Second place carried a cash prize of $ 160. As I vowed before
the tournament, this will be donated to the FSF.

Only four programs entered this year. Our big accomplishment
was winning the game against Smart Go, an old established program.
We won 2 out of three games in the round robin tournament. Many
Faces of Go was undefeated.

David Fotland mentioned the Free Software Foundation when he
handed me the envelope at the awards ceremony (I told him to), and
since I showed the program to a number of people at the Congress
or told them about it, there was a fair amount of publicity here
for the FSF and GNU Go.

Dan