[Computer-go] pro games againt pachi and mogo on KGS now

Olivier Teytaud olivier.teytaud at lri.fr
Thu Apr 14 04:32:23 PDT 2011


>
> h6 rengo against 5p+9p at 13:30(mogo) and 15:30(pachi) GMT+2.
>
>
Just to let you know: the Pachi game will for sure take place, but the mogo
game might be replaced by some unofficial experiments kindly accepted by the
pros (but we will not report this as it's unofficial and a last-minute
change and is just for analyzing some specific points interesting for us as
developers).

I can also give some other results from yesterday (consider this as
unofficial, we will write
something more carefully later):

a) 13x13
- mogo lost 2 games against the pros with H2 in 13x13
- mogo won 1 game over 2 against the pros with H2.5 in 13x13 (2.5 means: H3,
but some points given to white as komi, nearly equivalent to half a stone)

b) 9x9
 mogoTW won 3 out of 4 games in blind 9x9 Go. The pros said that for 9x9
blind-go makes
no difference for them. The opening book as black has been deeply modified.

c) 19x19

  We lost the games against the pros (H7 vs 9P, H6 vs 5P, rengo H6).

  We played games from random positions against a french 5D player (former
french champion and terrible for computers as a Monte-Carlo Go expert...).
We lost games with moderate number of stones in the initial position, but we
won 2 out of 3 games with 180 random stones whereas the human was allowed to
choose between black and white (for one position, the human believed it was
much easier for black, lost, and replayed the game with white, and lost
again). As this would never happen for standard games we can conclude that
random positions are a nightmare for humans. We will publish the positions.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20110414/b443ca2a/attachment.html>


More information about the Computer-go mailing list