[computer-go] SlugGo at the Cotsen Open

[David G Doshay <ddoshay@xxxxxxx>]

SlugGo was entered as a contestant in the Cotsen Open Go tournament in 
LA. This is the first time they allowed programs to enter just like 
people (except that people were allowed to refuse being paired against 
the computer; 3 people refused).

The day before the tournament I met Anders Kierulf and Bob Myers in 
West Hollywood. Anders started programming Go 20 years ago (it was his 
PhD thesis topic). Bob started the Intelligent Go Foundation, which is 
dedicated to computer Go. Anders and I set up the cluster in the 
driveway at Bob's condo because we did not want to try to lift the 
200-plus pound traveling cluster I had built over the previous month up 
to the the third floor. The cluster, 24 Mac minis and a dual G5 tower, 
draws more power than can be had from a single electrical circuit, so 
we dropped the 2 heavy power cords I had with me down from Bob's 
balcony. We got a few curious people asking what we were up to.

We played 2 games between his program SmartGo and SlugGo, and SlugGo 
won both. In the second game SlugGo gave SmartGo 4 handicap stones and 
still won by over 60 points! The game hinged on one bad move made by 
Smartgo. If it had chosen its second highest ranked move SmartGo 
probably would have won. But it was good for him because we found a bug 
in SmartGo that then allowed him to avoid the same mistake in the 
tournament (which would have resulted in a forfeit).

At the tournament SlugGo won the prize for the computer program that 
won the most games. The prize will be donated to the Intelligent Go 
Foundation. While it was not impressive that the number of wins was 
only one, Anders' program did not win any. SmartGo was entered as a 10 
kyu and SlugGo as a 9 kyu. I believe that our bracket was 8 to 12 kyu, 
and that is the range of players SlugGo faced. SlugGo's win was against 
an 8 kyu.

Two games that we lost were almost wins. In one game (against an 8 
kyu), we were very far ahead on points when the opponent started 
playing tactically complicated but clearly loosing moves and succeeded 
in making us loose on time. In the other game SlugGo (I think not the 
GNU Go code underneath) generated one mistake that made us loose. For 
reasons that are not yet clear, it made a ko in the corner, and that 
turned into the one extra eye that let an otherwise dead shape live. 
Almost won that one, just one wrong move. I think it was a horizon 
effect due to the shallow lookahead we needed to avoid loosing all our 
games on time. The other 2 games we lost were very reasonable and the 
other players deserved their wins. One loss was by only 10 points (to a 
10 kyu), the other by 28 (to a 12 kyu) and in that game the other 
player played very well. That player complimented SlugGo's ability to 
keep sente for several moves in a row. In all, I feel SlugGo played 
well, and I learned much.

We were also able to play a 9 stone handicap game against a 6 dan 
player who gave us commentary, so we can check the bad moves and figure 
out what we can do about them. He said SlugGo was clearly making a 
large number of good moves, a few excellent moves, and did seem to be 
much stronger than any other program he had played. He commented that 
SlugGo seemed "more natural" than the other programs he had played.

We got coverage from 2 Japanese language TV stations (one from Japan, 
one from LA), one of which interviewed me. I am supposed to get a copy 
of the tape.

As a result of this outing, I think it is likely that I will take the 
cluster to the US Go Congress in Tacoma WA in August.

I believe that the Cotsen will invite programs back next year, and I 
can recommend the experience.

Cheers,
David

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/