congratulations to mogo on its performance today!
it was an excellent result (1-2) versus a professional,
Thanks a lot. MoGo has also played some games against other
players (including a professional player) and I hope people
will accept that we publish the results (they are better than
against Catalin :-) ). We have also records of games against
very high level players a few weeks before the challenge.
Unfortunately, the cluster could not be used during the first game
and during the 19x19 game.
Catalin said that for the 9x9 games:
- mogo was very weak during the first game (the one
during which the cluster was replaced by a standard machine);
- Catalin had made a mistake in the second game (won by MoGo), and mogo
played quite well;
- the third game was interesting; Catalin won.
Catalin had been warned that it would be more difficult than a few years
ago - in particular he was aware of very good results against players of
professional level a few weeks ago. He has pointed out that the conditions
of the games were ok and that the time was sufficient for working
seriously.
According to Catalin, the komi 7.5 is too strong; he has been very
fair-play, saying that he would perhaps have lost 2 games instead of 1 if
the random choice win/black was different (all games were won by white).
In 19x19, Catalin won in spite of the 9 handicap stones. Catalin said
that MoGo was close to Dan level, as Catalin said it is usually easier for
him to win against a 1st Dan player with 9 stones than in that game
against MoGo. The cluster was
unavailable during the 19x19 game, seemingly due to a trouble in the
internet connection.
The won game is due in great part to the computer-go mailing list;
thanks a lot to all of you.
For the two games played by the cluster, it was 32 nodes of 8 cores, 3
GHz as far as I remember. The people providing the hardware can give more
details if they want to, I guess they read this mailing list.
I also want to point out that someone (probably reading this mailing
list) has claimed he has found a solution for winning against mogo,
not in the sense of a deterministic sequence but as a kind of situations
in which mogo is weak (as the Nakade trouble previously, kindly pointed
out by M. Fotland and others). I am unfortunately not able to understand
the situations,
I'll publish his comment if he can formalize it (well, if he read this
mailing list, he will perhaps do it himself :-) ).
We have removed most of the openings, because they have been generated
before we modify the behavior of mogo in front of Nakade, and the mogo
with new nakade-behavior seemingly does not like the openings generated
before the nakade-improvement. I guess a very
strong improvement will come easily, just by restarting the complete
(long and boring) process of generating opening books.
Best,
Olivier
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
