gnugo-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[gnugo-devel] arend_3_2.5: influence tuning: permeability


From: Arend Bayer
Subject: [gnugo-devel] arend_3_2.5: influence tuning: permeability
Date: Sun, 5 May 2002 17:07:34 +0200 (CEST)

 - decrease permeability reduction at diagonal neighbours of stones

Although this diff just adds one single character, its effects are big enough
to make it a separate patch... (Influence tuning is crazy!)
The hope is that it will make it less likely for GNU Go to leave bad holes
in connections when building territory/moyos, and to make necessary contact
plays like hanes etc. more attractive.

The regression delta is ~11 PASSes vs 3 FAILs: trevorb:300 is due to a
wrong connection analysis (which readconnect gets -- of course :) -- right),
and trevorb:600 due to not seeing a shortage-of-liberty owl problem.
(It might be worth an antisuji pattern, as it is a frequent post-joseki
position.)
The 3rd FAIL is the move in trevorc:600 which is a little worse than the
previous choice.

Arend


Index: engine/influence.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/influence.c,v
retrieving revision 1.54
diff -u -r1.54 influence.c
--- engine/influence.c  24 Apr 2002 11:54:36 -0000      1.54
+++ engine/influence.c  5 May 2002 13:26:32 -0000
@@ -1011,7 +1011,7 @@
          int dn = deltaj[k];
          if (ON_BOARD2(m+dm, n+dn) && q->p[m+dm][n+dn] == EMPTY) {
            /* Reduce less diagonally. */
-           float reduction = (k < 4) ? 0.25 : 0.5;
+           float reduction = (k < 4) ? 0.25 : 0.65;
            if (q->p[m][n] == BLACK)
              q->white_permeability[m+dm][n+dn] *= reduction;
            else




reply via email to

[Prev in Thread] Current Thread [Next in Thread]