Re: [gnugo-devel] look at regress file 13x13b:6

[Evan Berggren Daniel]

On Sun, 30 Mar 2003, kevin yong wrote:

> Hi, Evan & Gunnar:
>
>   thank you very much for you reply.
>
>   i did some work on 13x13b:6. following are what i
> found:
>
>   problem: gnugo should answer C4, but it did B4
> instead, which is wrong.
>
>   1. why/how B4 got to the top:
>
>   in the log file, it reports:
>
>     saving move for B6 found at B3!

This assessment is wrong, but it's not the fault of the tactics code.  The
tactics code understands the situation perfectly.  The owl code makes a
mistake here, I believe because it amalgamates B3 and B6 after B3 has been
played into the same dragon.  It is then able to save part of the dragon
(B3) and so reports success.  Finding out why this occurs and fixing it
would be reasonable.  It is probably the result of an owl pattern that
declares a connection when actually the connection reading code should be
consulted.

Using the sgf tracing to follow the owl code, discovering which pattern
is at fault, and correcting the pattern is probably the best approach
here.

>
>   then i trace to the source code, the dicision was
> made from reading.c do_find_defense()/defend2().
> However, when i insert a simple TRACE statement inside
> defend2(), when the str to be defended is B6, the log
> file reports hundreds of print of this TRACE. the call
> to defend2() originated from more than 6 places, and
> some of them indirectly recursive (such as do_attack()
> and attack4()). so, it is quite hard to pin-point
> exactly at which step the judegment was wrong. in
> other words, the conclusion came from a quite long and
> complicated process.
>   talk about solution, i think i need more detailed
> look into how the reading was done for this case.
>
> like to hear comments/suggestions from you.
>
> kevin.
>
>
>
>
>
> ______________________________________________________________________
> Post your free ad now! http://personals.yahoo.ca
>
>
> _______________________________________________
> gnugo-devel mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/gnugo-devel
>
>