Re: [gnugo-devel] Eye pattern 304

[Paul Pogonyshev]

Dan wrote:
> About the patch paul_3_20.5, which is the one discussed in
> this thread, I think the patch was supposed to address the
> position at move 62 in the game of S.P. Lee.
>
> But both before and after the patch, GNU Go currently
> wants to play at M6, not J2.
>
> J2 is valued 49.38 points and has this move reason:
>
>   J2: owl attack/defend for G2
>   J2: owl attack/defend for N3
>   ...
> M6 is valued 83.05 points:
>
>   M6: owl attack/defend for N3
>   M6: owl attack/defend for L5
>   ...
>
> ----------------------------------
>
> This is after Paul's patch. Before the patch:
>
>   J2: owl attack/defend for G2
>   ...
> (4.00) Move generation values J2 to 13.69
>
>   M6: owl attack/defend for L5
>   ...
> (0.00) Move generation values M6 to 55.99
>
> ----------------------------------
>
> Since GNU think's L5 is critical, this position should be
> considered a semeai. Of course W has much more space than
> B, so B should defend.
>
> I suppose we should use Paul's patch and also add this
> game as a regression.

i started the patch to fix the problem SP Lee found deep in owl reading
of the position. i never supposed it would fix this particular position
since there were more problems (like "getting deep, looks lively"). so
it is not a surprise it still prefers M6 (and even values it higher with
the patch). but it is good that owl_attack N3 doesn't return 0 anymore.
i believe the patch is correct.

and maybe someone could also make tests from the first game (unfinished)
SP Lee submitted. there were more mistakes in that one. but probably
some of them were fixed later with the semeai revision.

Paul