[Computer-go] codecentric Challenge 2016: Result of game 1

Nick Wedd maproom at gmail.com
Mon Aug 29 02:33:32 PDT 2016


On 29 August 2016 at 08:44, "Ingo Althöfer" <3-Hirn-Verlag at gmx.de> wrote:

> Hi,
>
> after a very interesting first game
> Lukas Kraemer scored a win against Zen19X.
>
> Some comments by Georg Snatzke (3-dan, from codecentric company)
> and sgf are here:
> https://go.codecentric.de/2016-game-1/
> https://go.codecentric.de/wp-content/uploads/2016/08/cgc2016A.sgf
>
> Zen19X was in the lead for a very long time.
> Most of the spectators started to realize around move 200
> that the table was turning. However, analysis with CrazyStone
> (edition 2016) showed that already around move 150 the
> distribution of outcomes of the pseudorandom games (within MCTS+)
> was no longer bell-shaped, but had large side peaks.
>
> Just two examples: after move 153, and after move 185
> http://www.dgob.de/yabbse/index.php?action=dlattach;
> topic=6212.0;attach=5157;image
> http://www.dgob.de/yabbse/index.php?action=dlattach;
> topic=6212.0;attach=5159;image


The large peak at the left of each graph, around 85 to the left of the main
peak, must show
the outcomes if White kills Black's O16 group.  But a kyu-player can see
that that group is
unconditionally alive: it has an eye at K19, with K17 and P14 as miai for a
second eye.  Do
you think that the presence of that peak had an effect on the result of the
game?

Nick

>
>
> So, Lukas had been able to create a position with a "whole-board" fight,
> where MCTS must have misled Zen19X.
>
> Game 2 of this "best of 5" match will be played on next Sunday.
>
> Ingo.
> _______________________________________________
> Computer-go mailing list
> Computer-go at computer-go.org
> http://computer-go.org/mailman/listinfo/computer-go




-- 
Nick Wedd      maproom at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20160829/8b28caf1/attachment.html>


More information about the Computer-go mailing list