[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnugo-devel] arend_1_29.2: more influence tuning
From: |
Arend Bayer |
Subject: |
[gnugo-devel] arend_1_29.2: more influence tuning |
Date: |
Tue, 26 Mar 2002 21:16:56 +0100 (CET) |
- assymmetry in influence computation reduced
- case added in strategic_penalty
- pattern tuning barriers.db and pattern.db
This is a pure tuning patch:
The reduced asymmetry comes from a reduction of the maximum value of
intrusions patterns in initial_influence and initial_opposite_influence.
This is to make opponent's moyo seem a little less secure and territory
and therefore make invasions a little more unattractive.
The pattern tuning is some influence pattern fixing that I hadn't done yet
plus some adjustments of shape values. Most notably, 3-3 invasions now
need smaller shape penalty.
The breakage is fine, see below.
The "contact fight" problem is not adressed by this patch.
Arend
3.1.28 + arend_1_29.1 + arend_1_29.2:
(excluding platform dependencies) 17 PASSes, 3 FAILs.
The two neurogo FAILs (pretty identical) are due to an owl failure.
./regress.sh . nngs1.tst
10 unexpected PASS!
16 unexpected PASS!
17 unexpected PASS!
34 unexpected PASS!
./regress.sh . neurogo.tst
8 unexpected FAIL: Correct 'O2', got 'Q5'
28 unexpected FAIL: Correct 'O2', got 'Q5'
./regress.sh . trevorb.tst
180 unexpected PASS!
250 unexpected PASS!
500 unexpected PASS!
830 unexpected PASS!
./regress.sh . manyfaces.tst
1 unexpected PASS!
./regress.sh . nngs.tst
120 unexpected PASS!
850 unexpected FAIL: Correct '!Q9|R9', got 'R9'
1060 unexpected PASS!
./regress.sh . trevorc.tst
280 unexpected PASS!
1040 unexpected PASS!
./regress.sh . arend.tst
38 unexpected PASS!
./regress.sh . 13x13.tst
12 unexpected PASS!
17 unexpected PASS!
(27 unexpected PASS!)
48 unexpected PASS!
./regress.sh . strategy4.tst
?? (not finished yet)
Index: engine/influence.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/influence.c,v
retrieving revision 1.44
diff -u -r1.44 influence.c
--- engine/influence.c 25 Mar 2002 15:25:27 -0000 1.44
+++ engine/influence.c 26 Mar 2002 20:01:01 -0000
@@ -523,7 +523,8 @@
/* Experimental influence: This function goes through the list of
* intrusion sources, and adds the intrusion as influence sources for color.
* The strength is corrected so that each stone's intrusions sources
- * can have total strength of at most the strength of the stone.
+ * can have total strength of at most 60%/100% of the strength of the stone.
+ * (100% is if q=&followup_influence, 60% otherwise).
*/
static void
add_marked_intrusions(struct influence_data *q, int color)
@@ -534,6 +535,7 @@
float strength_sum;
float correction;
float source_strength;
+ float allowed_strength;
gg_sort(q->intrusions, q->intrusion_counter, sizeof(q->intrusions[0]),
compare_intrusions);
@@ -567,8 +569,12 @@
|| q->intrusions[j].strength_pos != q->intrusions[j-1].strength_pos)
strength_sum += q->intrusions[j].strength;
}
- if (strength_sum > source_strength)
- correction = (source_strength / strength_sum);
+ if (q == &followup_influence)
+ allowed_strength = source_strength;
+ else
+ allowed_strength = 0.6 * source_strength;
+ if (strength_sum > allowed_strength)
+ correction = (allowed_strength / strength_sum);
else
correction = 1.0;
@@ -1758,8 +1764,8 @@
}
else
value_territory(&move_influence);
- if (m == (board_size - 19) + debug_influence_i
- && n == debug_influence_j && m >= 0) {
+ if (m == debug_influence_i
+ && n == debug_influence_j && m >= 0) {
if (printmoyo & PRINTMOYO_VALUE_TERRITORY)
print_influence_territory(&move_influence,
"territory (after move):\n");
Index: engine/value_moves.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/value_moves.c,v
retrieving revision 1.21
diff -u -r1.21 value_moves.c
--- engine/value_moves.c 25 Mar 2002 21:23:47 -0000 1.21
+++ engine/value_moves.c 26 Mar 2002 20:01:17 -0000
@@ -1046,6 +1046,7 @@
return adjusted_value;
}
+
/* The new (3.2) territorial evaluation overvalues moves creating a new
* group in the opponent's sphere of influence. The influence module cannot
* see that the opponent will gain by attacking the new (probably weak)
@@ -1081,6 +1082,17 @@
case EXPAND_MOYO_MOVE:
case STRATEGIC_ATTACK_MOVE:
continue;
+ /* If we find a tactical defense move, we just test whether it concerns
+ * a single-stone-dragon; if not, we stop, if yes, we let the necessary
+ * tests be made in the OWL_DEFEND_MOVE case.
+ */
+ case DEFEND_MOVE:
+ {
+ int target = worms[move_reasons[r].what];
+ if (dragon[target].size > 1)
+ return 0.0;
+ continue;
+ }
/* An owl defense of a single stone might be a stupid attempt to run
* away with an unimportant (kikashi like) stone. We assume this is the
* case if this single stone has a strong hostile direct neighbor.
@@ -1140,6 +1152,7 @@
ret_val = gg_max(0.0, ret_val);
return ret_val;
}
+
/*
* Estimate the direct territorial value of a move at (pos).
Index: patterns/barriers.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/barriers.db,v
retrieving revision 1.28
diff -u -r1.28 barriers.db
--- patterns/barriers.db 4 Mar 2002 06:49:08 -0000 1.28
+++ patterns/barriers.db 26 Mar 2002 20:01:20 -0000
@@ -869,6 +869,15 @@
:/,D
+Pattern Barrier62
+
+?xX?
+...O
+.!..
+
+:8,DOX
+
+
Pattern Intrusion1
xxxx
@@ -1034,6 +1043,12 @@
:8,B,value(30)
+?aX
+Xb!
+x..
+
+>return (!xplay_attack(b,a));
+
Pattern Intrusion7a
@@ -1232,11 +1247,12 @@
:8,B,value(30)
-oQo
-aXO
+oeo
+aXd
cbX
;!oplay_attack(a,b,c,a)
+>return (lib(d)>2 && lib(e)>2 && !xplay_attack(a,e));
Pattern Intrusion16
@@ -1677,7 +1693,7 @@
;safe_omove(a)
;&& (o_somewhere(e) || oplay_attack(a,e,e))
;&& !oplay_defend_both(a,b,c,b,D)
->return (!xplay_attack(e,f));
+>return (!xplay_attack(a,f) && (o_somewhere(e) || (!xplay_attack(e,f))));
Pattern Intrusion45
Index: patterns/fuseki.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/fuseki.db,v
retrieving revision 1.12
diff -u -r1.12 fuseki.db
--- patterns/fuseki.db 20 Mar 2002 15:02:22 -0000 1.12
+++ patterns/fuseki.db 26 Mar 2002 20:01:23 -0000
@@ -207,7 +207,7 @@
|.....
+-----
-:/,Feb,shape(-6)
+:/,Feb,shape(-2)
Pattern F8
Index: patterns/patterns.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/patterns.db,v
retrieving revision 1.60
diff -u -r1.60 patterns.db
--- patterns/patterns.db 25 Mar 2002 21:23:47 -0000 1.60
+++ patterns/patterns.db 26 Mar 2002 20:01:40 -0000
@@ -8868,13 +8868,14 @@
Pattern CB327
# gf Reduced shape value. (3.1.23)
# See nngs:1750
+# ab increased shape value (3.1.29)
?.? seal opponent in
...
.*O
OX?
-:8,OXcbeEd,shape(2)
+:8,OXcbeEd,shape(3)
Pattern CB328
@@ -10520,6 +10521,7 @@
Pattern ED90
# db New pattern (3.1.28)
+# ab d classification added (3.1.29)
???..o? Jump under on the third line
???X.oo
@@ -10528,7 +10531,7 @@
o......
-------
-:8,j
+:8,jd
???..o?
???X.oo
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [gnugo-devel] arend_1_29.2: more influence tuning,
Arend Bayer <=