A couple of hours.
/Gunnar
Michael Williams wrote:
Very cool. How long has this been going on?
Gunnar Farnebäck wrote:
At http://trac.gnugo.org/6x6.sgf you can find an ongoing analysis of
6x6. This is a very big and quite raw sgf file where each node has a
comment block looking like this:
00000000000000000
-2.5: 0 4 black
-0.5: 5 9 black
1.5: 9 10 black
3.5: 9 34 white
5.5: 38 20 white
7.5: 22 7 white
9.5: 6 0 white
B3 4b47e72a68d23f011 +
B4 4b47e72a68d23f011 +
[...]
F5 00f9e9a88c0120581
F6 01f9ba0f1b5c68e11
PASS 00000000000000001
The first line is the id of the position in the form of a symmetry
invariant 64-bit Zobrist board hash, written in hexadecimal,
concatenated with the move number.
Following are the results for different komi values. Listed on each
line is komi value, number of observed white wins, number of observed
black wins, and minimax winner.
Last comes a list of moves and id of the obtained positions. If the id
is followed by a plus sign it means that move has been played and can
be found as a child node. If it is followed by a minus sign the move
has been played but gives a transposition of a a position that has
been expanded elsewhere in the tree.
The example above is from the root node and means that correct komi
currently is supposed to be between 1.5 and 3.5 (i.e. 2 or possibly,
but unlikely, 3). This may very well have changed by the time you
download the sgf.
So far this analysis has only been done by GNU Go, with and without
the --monte-carlo option. I would like other, stronger programs to
join in.
To do that, just point your regular cgos client to trac.gnugo.org,
port 6867. This is of an experimental nature and hopefully works for
more people than me. Don't worry about user names and passwords, those
are not used for anything yet, just fill in something. Time controls
are not used and the server will always report five minutes remaining
time. Interrupted games are lost forever, but that's not really a
problem.
If you know how to play manually on cgos, feel free to do so here.
Btw, programs are not playing against each other, but against the
server following supposedly winning lines, falling back on GNU Go when
it finds an unseen position.
/Gunnar
_______________________________________________
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/
