Re: [gnugo-devel] Ladder attack overlooked?

[Evan Berggren Daniel]

To get some details on what is going on, you can load the sgf file in GTP
mode.

$gnugo --mode gtp --quit
loadsgf 44.sgf 24

That will load the file up to but not including move 24.

gg_genmove black

That will return gnugo's best move for black.  In this case, S17.

top_moves_black

will return a list of the top moves and their valuations.  We see S17
valued at 24.91, and H14 valued at 19.51.

attack J17

tries to tactically attack (nets, ladders, etc as opposed to
eyespace-reduction) the group at J17.  We see that gnugo finds the attack
at H14.

My interpretation as to what is going on is that gnugo values capturing
the 5 stones at about 20 points; this seems reasonable, though perhaps a
little low.  However, it values S17 at 25 points, which looks far too high
to me.  The other problem is that is values H11, F11 and F12 slightly
higher than H14.  Presumably it thinks those moves keep the w stones dead,
but have greater influence on the rest of the game.

If you want details, you can look at the output of gnugo -l 44.sgf -L 24
-t

Interesting problem.

Thanks

Evan Daniel

On Sun, 30 Mar 2003, SP LEE wrote:

>
> Hi,
>
> In the attached game, GNUGO seemed to have overlooked the move at H14 to
> catch the 5 white stones in a ladder attack. I have tried to trace the
> reasoning of GNUGO, but so far I don't understand fully what GNUGO thought.
> Is the function 'simple_ladder_attack' involved in this case?
>
> Best regards,
>
> SP Lee
>
>
>
>
> _________________________________________________________________
> MSN 8 helps eliminate e-mail viruses. Get 2 months FREE*.
> http://join.msn.com/?page=features/virus
>