From 04fd7e044476eabc94a3c0fa8f7326be317b0337 Mon Sep 17 00:00:00 2001
From: Petr Baudis <pasky@ucw.cz>
Date: Fri, 2 Oct 2009 16:54:53 +0200
Subject: [PATCH] playoutp_assess(): Iterate over prior map within policy
 itself

---
 playout.h       |  9 ++++++---
 playout/moggy.c | 19 ++++++++++++++++++-
 uct/prior.c     | 14 ++------------
 3 files changed, 26 insertions(+), 16 deletions(-)

diff --git a/playout.h b/playout.h
index 92f4db3..179110e 100644
--- a/playout.h
+++ b/playout.h
@@ -4,13 +4,16 @@
 struct board;
 struct move;
 enum stone;
+struct prior_map;
 
 
 struct playout_policy;
 typedef coord_t (*playoutp_choose)(struct playout_policy *playout_policy, struct board *b, enum stone to_play);
-/* number of won (>0) or lost (<0) games to seed the node with (usually a proportion of @games);
- * can return 0 if policy has no opinion */
-typedef int (*playoutp_assess)(struct playout_policy *playout_policy, struct board *b, struct move *m, int games);
+/* Set number of won (>0) or lost (<0) games for each considerable
+ * move (usually a proportion of @games); can leave some untouched
+ * if policy has no opinion. The number must have proper parity;
+ * just use uct/prior.h:add_prior_value(). */
+typedef void (*playoutp_assess)(struct playout_policy *playout_policy, struct prior_map *map, int games);
 typedef bool (*playoutp_permit)(struct playout_policy *playout_policy, struct board *b, struct move *m);
 
 struct playout_policy {
diff --git a/playout/moggy.c b/playout/moggy.c
index 008a10b..191275e 100644
--- a/playout/moggy.c
+++ b/playout/moggy.c
@@ -11,6 +11,7 @@
 #include "playout/moggy.h"
 #include "random.h"
 #include "tactics.h"
+#include "uct/prior.h"
 
 #define PLDEBUGL(n) DEBUGL_(p->debug_level, n)
 
@@ -699,7 +700,7 @@ assess_local_bonus(struct playout_policy *p, struct board *board, struct move *a
 }
 
 int
-playout_moggy_assess(struct playout_policy *p, struct board *b, struct move *m, int games)
+playout_moggy_assess_one(struct playout_policy *p, struct board *b, struct move *m, int games)
 {
 	struct moggy_policy *pp = p->data;
 
@@ -788,6 +789,22 @@ playout_moggy_assess(struct playout_policy *p, struct board *b, struct move *m,
 	return 0;
 }
 
+void
+playout_moggy_assess(struct playout_policy *p, struct prior_map *map, int games)
+{
+	/* TODO: Optimize this! */
+	foreach_point_and_pass(map->b) {
+		if (!map->consider[c])
+			continue;
+		int assess = 0;
+		struct move m = { c, map->to_play };
+		assess = playout_moggy_assess_one(p, map->b, &m, games);
+		if (!assess)
+			continue;
+		add_prior_value(map, c, assess, abs(assess));
+	} foreach_point_end;
+}
+
 bool
 playout_moggy_permit(struct playout_policy *p, struct board *b, struct move *m)
 {
diff --git a/uct/prior.c b/uct/prior.c
index 89b2a94..de470a0 100644
--- a/uct/prior.c
+++ b/uct/prior.c
@@ -96,18 +96,8 @@ void
 uct_prior_playout(struct uct *u, struct tree_node *node, struct prior_map *map)
 {
 	/* Q_{playout-policy} */
-	foreach_point_and_pass(map->b) {
-		if (!map->consider[c])
-			continue;
-		int assess = 0;
-		if (u->playout->assess) {
-			struct move m = { c, map->to_play };
-			assess = u->playout->assess(u->playout, map->b, &m, u->policy_eqex);
-		}
-		if (!assess)
-			continue;
-		add_prior_value(map, c, assess, abs(assess));
-	} foreach_point_end;
+	if (u->playout->assess)
+		u->playout->assess(u->playout, map, u->policy_eqex);
 }
 
 void
-- 
2.11.4.GIT