[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] eye patterns
From: |
Gunnar Farneback |
Subject: |
Re: [gnugo-devel] eye patterns |
Date: |
Tue, 29 Apr 2003 19:27:42 +0200 |
User-agent: |
EMH/1.14.1 SEMI/1.14.3 (Ushinoya) FLIM/1.14.2 (Yagi-Nishiguchi) APEL/10.3 Emacs/20.7 (sparc-sun-solaris2.7) (with unibyte mode) |
Paul wrote:
> this patch is extracted from paul_3_18.7. regression delta is negative,
> but i believe the patch itself is correct.
>
> strategy2:77 FAIL H15 [H3]
> strategy2:80 FAIL S8 [P4|Q4|Q3]
> nngs3:740 FAIL F2 [!F2]
>
>
> strategy2:77 now gnu go prefers H15 since it thinks H15 attacks H17. the patch
> saves some owl nodes here: original gnu go with OWL_NODE_LIMIT
> set to 1200 says H15 as well.
This test is revised by one of Evan's pending patches and should no
longer be a problem.
> strategy2:80 S8 looks like a correct move to me (in addition to P4, Q4 and
> Q3).
> it's not that solid, but it does win the semeai. i don't exactly
> understand why it doesn't like Q4 anymore, owl_does_attack Q4 R5
> says "1". however, nonpatched gnu go with OWL_NODE_LIMIT raised
> to 2000 agrees with patched one that S8 is better than Q4.
S8 leaves significant aji and loses points compared to the connecting
moves so I wouldn't consider it as correct. But the choice of moves is
a move valuation issue and unrelated to this patch.
> nngs3:740 gnu go was never good in reading such unclear positions. with
> the patch it sees that E2 doesn't owl defend C2 (because it
> orders attacking moves differently and doesn't run out of nodes).
> so it devalues E2 a couple points and F2 becomes the top move.
> nonpatched gnu go with OWL_NODE_LIMIT == 2000 doesn't count E2
> as an owl defence for C2 (it doesn't like F2 at all, but that's
> another story).
Seems to be passing rather much by luck currently.
Most of the pattern changes look correct. A few comments though:
> @@ -661,19 +660,13 @@ Pattern 469
>
> Pattern 491
>
> - .
> -!X!
> -
> -:0001
> -
> -Pattern 492
> -
Anything wrong with that pattern?
> @@ -2783,13 +2776,14 @@ x..x@
> Pattern 6261
> # tm New Pattern (3.1.22) (see owl:111)
>
> - @
> -!.*.!
> + )
> address@hidden>.@
>
> :0011
I don't quite get this one.
> Pattern 6262
> +# FIXME: I don't understand this pattern. Is it correct? -pp
>
> !
> !x..!
Agreed, looks strange.
> @@ -4399,6 +4393,8 @@ Pattern 70004
>
>
> Pattern 70005
> +# FIXME: This pattern looks like 1112 to me. "$...<." seems to be
> +# attackable at '<'. -pp
>
> $.....@
Also agreed. The "$>..*." pattern should be added too.
/Gunnar