I have been downloading a lot of games from CGOS recently to get some
kind of picture of how the opening moves are played for 9x9 by
programs rated about 2200+ but it is really hard to say anything.
After a few moves there is not enough data to make any conclusions.
And Early on where there are data, you have the problem that the
subtree might distort the root you are looking at because a weaker
program always play and loses with a certain move.
With selfplay it is easier to get more data which is easier to
interpret because all player are equally strong but then biases of the
program will make conclusion about which moves are good really hard.
Also it seems to me that finding close to unbeatable play for 9x9 is
really really hard (if anyone doubted that). There seems to be up to
roughly 3-8 playable candidate moves in almost all opening positions
in the variations I have been looking at.
Valkyria currently play quite narrow book starting with E5. Since the
last tournament I noticed I should try to break symmetry before
letting it think because other wise it could spend a minute on two
moves that are symmetric. But I am considering to widen the choices
for Black on the first move.
-Magnus (who finally was able to register with an email account which
only put this mailing list in the spam folder rather than just delete
it)
Quoting Martin Mueller <[email protected]>:
It is just an experiment that I am running right now. I don't know
which move is better. The M in the version number 1111M indicates
that the program is a modified version. In this case the
modification is that I added or changed a few moves in the book. But
I am not confident in these changes so it is not in the svn. This
modified version will randomly choose either E5 or E4. See the diff
below.
Martin
diff book.dat book-3.dat
1504c1504
< 9 | E5
---
9 | E5 E4
1698c1698
< 9 E5 G5 | F3 G4
---
9 E5 G5 | F3
2370c2370
< 9 F5 D5 E6 | F4
---
9 F5 D5 E6 | E4
2732c2732
< 9 D5 F5 F4 G4 E4 G3 F6 | G6
---
9 D5 F5 F4 G4 E4 G3 F6 | G5
2913a2914,2919
9 E5 E3 C4 C3 B3 D4 | D5
9 E5 E3 C4 D4 | D5
9 E5 E3 C4 D4 D5 C3 | E4
9 E5 E3 C4 D4 D5 C3 E4 D3 | G4
9 E4 E6 | D6
9 E4 E6 D6 D7 D5 C7 | F6
Martin
I noticed on CGOS that the latest Fuego is starting at the 5-4 point rather
than the 5-5 point. I know that Fuego uses a book. Does this mean that
analysis of games between strong programs shows that 5-4 is significantly
better than 5-5?
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
