We tried a set of 3x3 patterns that were culled from a set of cgos
games involving the best programs. We did not have much success in
using them as predictors of eventual winner. That is not to say that
they can serve no purpose, but when we had such low success in win
prediction we felt that they did not meet a pretty low hurdle, and
are not looking at them any more.
Cheers,
David
On 18, Sep 2007, at 1:58 PM, Jason House wrote:
Since I've started thinking about adding 3x3 patterns in my own
code, has anything ever evolved from this? For example, I'd be
interested in contrasting the frequency of playing a pattern with
the frequency that the opportunity to play the pattern pops up.
On 5/26/07, Peter Drake <[EMAIL PROTECTED]> wrote:
I've thrown together a rudimentary pattern extractor and was
looking at 3x3 patterns taken from a collection of 9x9 games
provided by Nici Schraudolph. Each game was played through
in each of the four rotations and two reflections. Colors were
inverted every other move, so all of these patterns have black to
play in the center.
Looking at the data below, it appears that:
a) It might be better to somehow take into account the ratio
between how often the pattern appeared and how often the
corresponding move was played.
b) Endgame moves appear often, because they get played sooner or
later.
c) Tesuji don't appear as often as one would hope. Connecting
against a peep, making a bamboo joint, and the diagonal cut are not
among the top patterns.
Let the thread begin!
The actual information:
134008 occurrences of
...
.?.
...
(Fair enough, you need a lot of these.)
9794 occurrences of
...
.?.
Bw.
(And similarly for the other 7 transformations of this. Hane.)
7444 occurrences of
...
.?.
.w.
(3 transformations. Contact play or attachment.)
7170 occurrences of
...
.?.
w..
(3 transformations. Shoulder hit.)
5512 occurrences of
###
.?.
Bw.
(7 transformations. Hane underneath -- is there another term?)
4840 occurrences of
...
w?.
Bw.
(3 transformations. Crosscut.)
4489 occurrences of
...
w?.
wB.
(7 transformations. Is there a name for this?)
4054 occurrences of
...
B?.
wB.
(3 transformations. Preventing a crosscut? Closed triangle?)
3922 occurrences of
...
.?.
wB.
(7 transformations. Anti-hane?)
3806 occurrences of
...
.?.
B..
(3 transformations. Diagonal connection.)
3656 occurrences of
###
.?.
BwB
(3 transformations. Connection underneath.)
3597 occurrences of
...
w?.
.B.
(7 transformations. Name?)
3197 occurrences of
...
w?.
BB.
(7 transformations. Another full triangle.)
3196 occurrences of
..#
.?#
Bw#
(7 transformations. Blocking on the first line -- VERY endgame.)
2986 occurrences of
###
B?.
wB.
(7. Connecting on the first line?)
2892 occurrences of
...
.?.
BwB
(3. Potentially a capture.)
2456 occurrences of
...
.?.
wBw
(3. The opposite of the previous pattern.)
2300 occurrences of
###
.?.
Bww
(7. Name?)
Peter Drake
http://www.lclark.edu/~drake/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
