[gnugo-devel] arend_1_30.3: additional ko value

[Arend Bayer]

 - ko threat value computation changed
 
It seems to me that the old computation of move[pos].additional_ko_value
in value_moves.c is flawed. If a move has reverse_followup_value, but
no followup value, then it got previously awarded a threat value.

This patch changes this. It did not change any regression, so this
was apparently not an urgent problem, although I vaguely remember seeing
it somewhere a few releases ago.

Arend


Index: engine/value_moves.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/value_moves.c,v
retrieving revision 1.22
diff -u -r1.22 value_moves.c
--- engine/value_moves.c        26 Mar 2002 21:32:59 -0000      1.22
+++ engine/value_moves.c        29 Mar 2002 01:41:41 -0000
@@ -2218,14 +2218,13 @@
               move[pos].reverse_followup_value);
       }
 
-      /* If a ko fight is going on, we should use the full followup
-       * and reverse followup values in the total value. We save the
-       * additional contribution for later access.
+      /* If a ko fight is going on, we count sente moves with their
+       * full value. We save the additional contribution for later access.
        */
       move[pos].additional_ko_value =
-       followup_value 
-       + move[pos].reverse_followup_value 
-       - (tot_value - old_tot_value);
+          gg_min(0.5 * followup_value + 0.5 * move[pos].reverse_followup_value,
+                1.0 * tot_value + followup_value)
+         - (tot_value - old_tot_value);
 
       /* Not sure whether this could happen, but check for safety. */
       if (move[pos].additional_ko_value < 0.0)
@@ -2233,8 +2232,9 @@
     }
     else {
       move[pos].additional_ko_value =
-       shape_factor * (move[pos].followup_value
-                       + move[pos].reverse_followup_value);
+         shape_factor * gg_min(0.5 * followup_value
+                                 + 0.5 * move[pos].reverse_followup_value,
+                               followup_value);
     }
 
     tot_value += 0.05 * move[pos].secondary_value;
@@ -2273,9 +2273,9 @@
   }
   else {
     move[pos].additional_ko_value =
-      shape_factor * (move[pos].followup_value +
-                     gg_min(move[pos].followup_value,
-                            move[pos].reverse_followup_value));
+         shape_factor * gg_min(0.5 * move[pos].followup_value
+                                 + 0.5 * move[pos].reverse_followup_value,
+                               move[pos].followup_value);
   }
 
   /* If the move is valued 0 or small, but has followup values and is