[gnugo-devel] count territory change for atari_atari defense moves

[Gunnar Farnebäck]

This patch forces territory evaluation for moves defending against a
combination attack.

nngs:1060       FAIL T3 [H3|F3]
strategy3:116   FAIL O7 [M4|N5]
nngs3:1090      PASS N9 [R7|N9|M8]
gunnar:59       pass
2 PASS (1 PASS, 1 pass)
2 FAIL
Total nodes: 1664503076 3063412 13241195 (+0.17% -3.3e-05% +0.17%)

nngs:1060 is accidental, due to a bogus combination attack. See the
recently added atari_atari:28.

strategy3:116 can also be considered accidental in that it uncovers a
different move valuation problem. There's nothing unreasonable about
the changes introduced by the patch.

nngs3:1090 has been discussed recently and is not all that
interesting. Still the patch gives a minor improvement.

gunnar:59 is the targeted testcase, where this patch correctly chooses
an atari_atari defense on a dame point instead of inside own
territory.

- atari_atari moves considered blocking in estimate_territorial_value().

/Gunnar

Index: engine/value_moves.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/value_moves.c,v
retrieving revision 1.131
diff -u -r1.131 value_moves.c
--- engine/value_moves.c        13 Sep 2004 10:45:11 -0000      1.131
+++ engine/value_moves.c        22 Sep 2004 15:39:55 -0000
@@ -2323,6 +2327,11 @@
       break;
       
     case YOUR_ATARI_ATARI_MOVE:
+      /* Set does_block to force territorial valuation of the move.
+       * That way we can prefer defenses against combination attacks
+       * on dame points instead of inside territory.
+       */
+      does_block = 1;
       this_value = move_reasons[r].what;
       tot_value += this_value;
       TRACE("  %1m: %f - defends against combination attack on several 
worms\n",