[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [computer-go] SlugGo vs Many Faces, newest data
I'm not a statician, but those are a fair number of games already, and
everything seems to indicate about 4 stones over MFG.
I think this is the biggest leap ahead in computer-Go since the late '80s,
congratulations!
Maybe it's not the most efficient use of computer power, but at least it's a
significant improvement. This in contrast to the rather lack-luster
improvements of recent years despite the ever-increasing performance of PCs.
Hopfully this will point to new developments to improve computer-go.
Mark Boon
-----Original Message-----
From: computer-go-bounces@xxxxxxxxxxxxxxxxx
[mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx]On Behalf Of David G Doshay
Sent: Saturday, September 04, 2004 4:41
To: computer-go
Subject: [computer-go] SlugGo vs Many Faces, newest data
In our first set against Many Faces of Go, where we played White with
no handicap and no komi and the results were:
> Black wins 5 (by: 33, 15, 75, 4, 21)
> White wins 20 (by: 126, 59, 67, 73, 40, 17, 25, 76, 58, 21, 25, 125,
> 24, 2, 42, 54, 41, 1, 32, 53)
After I posted the above set Doug Ridgway suggested we start giving
Many Faces a handicap, so we ran some 2, 3, and 4 stone games.
Through 28 August we generated the following results. SlugGo is always
white against Many Faces of Go playing black, no komi.
Handicap = 2
42 games
Black won 11 (by: 11, 3, 45, 66, 31, 36, 3, 17, 27, 7, 3)
White won 31( by: 33, 31, 40, 2, 25, 23, 9, 56, 57, 17, 14, 21, 65,
37, 77, 16, 91, 51, 78, 15, 4, 2, 74, 6, 35, 23, 30, 18, 9, 31, 51)
Handicap = 3
6 games
Black won 4 (by: 25, 135, 30, 107)
White won 2(by: 21, 51)
Handicap = 4
18 games
Black won 8 (by: 11, 36, 58, 15, 12, 6, 13, 56)
White won 9 ( by: 11, 104, 79, 49, 20, 72, 51, 24, 10)
One tie
I sent the above to Doug, to which he replied:
> There's a plot of the data, best
> fit line, and 95% CI of score predictions at
> http://dridgway.com/Go/sluggo_vs_MFG.png .
Doug also suggested we try higher handicap games. We did,
although we did not collect very many.
Newest results since we reported the above to Doug are:
Handicap = 3
25 games
SlugGo wins 13 (by: 41, 20, 25, 50, 15, 23, 8, 61, 49, 66, 17, 9, 11)
MFG wins 12 (by: 39, 4, 14, 17, 74, 136, 73, 52, 61, 88, 2, 29)
Handicap = 4
27 games
SlugGo wins 14 (by: 1, 3, 76, 17, 32, 22, 1, 16, 48, 29, 17, 37, 91, 68)
MFG wins 13 (by: 19, 39, 43, 107, 26, 57, 21, 16, 22, 16, 3, 66, 93)
I found it interesting that Many Faces won 10 of the first 13, and then
SlugGo won 11 of the next 14. This was what I meant in an earlier
email when I wrote about how easy it is to be fooled by small samples.
It looks to me like we would need at least 200 games to know what
is going on at 4 stones. All of these games involved huge groups and
large strings of cut off stones.
Handicap = 5
6 games
SlugGo wins 2 (by: 1, 31)
MFG wins 4 (by: 15, 33, 17, 8)
Handicap = 6
5 games
SlugGo wins 2 (by: 13, 36)
MFG wins 3 (by: 24, 5, 7)
I did not expect to win any 6 stone games, so I was pleased. But as said
earlier, these are very small samples and I do not trust small samples.
Perhaps Doug will be so kind as to add the new data to the other and
send along a pointer to a plot of the full data set.
This will be the last of the data from playing SlugGo against Many Faces
for a while. We are going to concentrate on cleaning up the code and
changing the logic in a few ways. I do not plan to play any more games
with the present code, but will start again fresh with the new version.
Cheers,
David
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/