[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] nasty shape values
From: |
Gunnar Farneback |
Subject: |
Re: [gnugo-devel] nasty shape values |
Date: |
Wed, 26 Feb 2003 22:35:15 +0100 |
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) |
Nando wrote:
> > (...) the mistake was caused by CB328 pattern with shape value of +10.
>
> FWIW, I also have a (not submitted) test case where CB328 causes a mistake
> worth about 10 points. There has been a discussion about high shape values
> some time ago, with the result that nothing changed. Still, shape(10)
> roughly means doubling the move value whatever this value is, which
> looks a bit unreasonable to me.
Please submit such tests.
As an experiment I tried to significantly limit the effects of shape
values to a maximum change of 0.95 points, as implemented in the
appended patch (not for CVS). The breakage is large and more negative
than positive. Analysis of these tests may give some insights about
the positive and negative effects of (large) shape values.
trevora:130 FAIL B7 [F4]
nngs1:20 FAIL G3 [F17]
nngs1:34 PASS N15 [!N10|N9|N11|O11|O10]
nngs1:35 FAIL T6 [D13]
nngs1:48 PASS R16 [R16]
golife:4 FAIL B4 [H5]
lazarus:16 PASS L16 [L16]
strategy2:51 FAIL O7 [P3]
strategy2:53 FAIL E10 [S15]
strategy2:54 FAIL P7 [P6]
strategy2:70 PASS N16 [M15|J15|N16|J16]
strategy2:84 PASS Q13 [Q13]
strategy2:90 FAIL F5 [B14|D17]
nicklas1:501 FAIL F6 [G7]
manyfaces:2 FAIL T15 [S16]
trevor:220 FAIL A7 [E8]
trevor:1110 PASS O12 [O12]
nngs:200 PASS O11 [P12|O11]
nngs:590 PASS G3 [G3]
nngs:820 FAIL D14 [J13|L9]
nngs:910 PASS T16 [T16]
nngs:950 FAIL C8 [F1]
nngs:1050 PASS G10 [J10|H10|G10|F10]
trevorc:1570 FAIL E1 [E2|F2|F3]
strategy3:128 FAIL R11 [O8]
global:9 PASS F4 [F4]
arend:38 FAIL O17 [O5]
13x13:12 FAIL H6 [G7|F7|G8]
13x13:36 FAIL B2 [C5]
13x13:37 FAIL E3 [E10]
trevord:200 PASS O11 [O11|E6]
nngs3:310 PASS B5 [C5|B5]
nngs3:470 PASS P11 [!J17]
nngs3:800 FAIL E3 [G4]
nngs4:30 FAIL D5 [C5|C6]
strategy5:224 FAIL O18 [N15]
strategy5:234 FAIL D16 [H14]
century2002:150 PASS B18 [B18]
safety:5 FAIL C15 [K8|L9|L7]
gunnar:19 PASS L4 [L4]
13x13b:22 FAIL A6 [B6|B7]
13x13b:40 FAIL H10 [F8]
13x13b:43 PASS J3 [J3]
/Gunnar
Index: engine/value_moves.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/value_moves.c,v
retrieving revision 1.82
diff -u -r1.82 value_moves.c
--- engine/value_moves.c 13 Feb 2003 20:09:42 -0000 1.82
+++ engine/value_moves.c 26 Feb 2003 21:21:57 -0000
@@ -2590,6 +2590,10 @@
/* shape_factor has already been computed. */
float old_value = tot_value;
tot_value *= shape_factor;
+ if (tot_value > old_value)
+ tot_value = old_value + soft_cap(tot_value - old_value, 0.95);
+ else
+ tot_value = old_value - soft_cap(old_value - tot_value, 0.95);
if (verbose) {
/* Should all have been TRACE, except we want field sizes. */
gprintf(" %1m: %f - shape ", pos, tot_value - old_value);