Re: [gnugo-devel] look at regress file 13x13:4

[Evan Berggren Daniel]

On Wed, 2 Apr 2003, kevin yong wrote:

> Hi, Evan:
>
>   thank you for your reply.
>
>   i did some work on 13x13:4. Problem: gnugo should
> choose L9 instead of H8.
>
>   after exam the log file, i found that: the dragon J8
> contains only 6 stones, but not including J10 & L8. i
> think they should all be in one dragon.

All one dragon is dangerous.  Obviously, J10 and L8 are not connected to
each other.  Since the connection is so tenuous, I would be tempted to
think the current amalgamation is reasonable.

>
>   then, i did a simple test: somehow i lump J10 & L8
> into J8, and all becomes a single dragon. then gnugo
> totally ingnore the dragon, and choose to move at F13.
>
>   this raise a important issue, before we exam the
> detail of which move should be choosen: dragon should
> be correctly lumped before generate moves. obviously,
> different dragon formation give out different gen-move
> results!
>
>   this suggests that, before we work on issue of
> H8/L9/F13, we need to look at the dragon
> build/formation first.
>
>   like to hear your comments/suggestions.
>
> kevin.

Have you looked at what moves are tried as attacks and defenses?  what
moves are missing from that consideration?  Usually these problems arise
from gnugo not seeing the relevant move.  After you have identified the
move(s) in question, you can add or modify owl patterns to suggest those
moves.

Evan Daniel