4 #define MAX_GAMELEN 600
10 struct board_ownermap
;
13 /** Playout policy interface: */
15 struct playout_policy
;
16 /* Initialize policy data structures for new playout; subsequent choose calls
17 * (but not assess/permit calls!) will all be made on the same board; if
18 * setboard is used, it is guaranteed that choose will pick all moves played
19 * on the board subsequently. The routine is expected to initialize b->ps
20 * with internal data. At the playout end, b->ps will be simply free()d,
21 * so make sure all data is within single allocated block. */
22 typedef void (*playoutp_setboard
)(struct playout_policy
*playout_policy
, struct board
*b
);
23 /* Pick the next playout simulation move. */
24 typedef coord_t (*playoutp_choose
)(struct playout_policy
*playout_policy
, struct board
*b
, enum stone to_play
);
25 /* Set number of won (>0) or lost (<0) games for each considerable
26 * move (usually a proportion of @games); can leave some untouched
27 * if policy has no opinion. The number must have proper parity;
28 * just use uct/prior.h:add_prior_value(). */
29 typedef void (*playoutp_assess
)(struct playout_policy
*playout_policy
, struct prior_map
*map
, int games
);
30 /* Allow play of randomly selected move. */
31 typedef bool (*playoutp_permit
)(struct playout_policy
*playout_policy
, struct board
*b
, struct move
*m
);
32 /* Tear down the policy state; policy and policy->data will be free()d by caller. */
33 typedef void (*playoutp_done
)(struct playout_policy
*playout_policy
);
35 struct playout_policy
{
37 /* We call setboard when we start new playout.
38 * We call choose when we ask policy about next move.
39 * We call assess when we ask policy about how good given move is.
40 * We call permit when we ask policy if we can make a randomly chosen move. */
41 playoutp_setboard setboard
;
42 playoutp_choose choose
;
43 playoutp_assess assess
;
44 playoutp_permit permit
;
46 /* Particular playout policy's internal data. */
51 /** Playout engine interface: */
53 struct playout_setup
{
54 int gamelen
; /* Maximal # of moves in playout. */
56 /* XXX: We used to have more, perhaps we will again have more
61 struct playout_amafmap
{
62 /* Record of the random playout - for each intersection:
63 * S_NONE: This move was never played
64 * S_BLACK: This move was played by black first
65 * S_WHITE: This move was played by white first
67 enum stone
*map
; // [board_size2(b)]
69 /* the lowest &0xf is the enum stone, upper bits are nakade
70 * counter - in case of nakade, we record only color of the
71 * first stone played inside, but count further throwins
72 * and ignore AMAF value after these. */
73 #define amaf_nakade(item_) (item_ >> 8)
74 #define amaf_op(item_, op_) do { \
76 item_ = (mi_ & 0xf) | ((amaf_nakade(mi_) op_ 1) << 8); \
79 /* Additionally, we keep record of the game so that we can
80 * examine nakade moves; really going out of our way to
81 * implement nakade AMAF properly turns out to be crucial
82 * when reading some tactical positions in depth (even if
83 * they are just one-stone-snapback). */
84 struct move game
[MAX_GAMELEN
+ 1];
86 /* Our current position in the game sequence; in AMAF, we search
87 * the range [game_baselen, gamelen]. */
90 /* Whether to record the nakade moves (true) or just completely
91 * ignore them (false; just the first color on the intersection
92 * is stored in the map, nakade counter is not incremented; game
93 * record is still kept). */
98 /* >0: starting_color wins, <0: starting_color loses; the actual
99 * number is a DOUBLE of the score difference
100 * 0: superko inside the game tree (XXX: jigo not handled) */
101 int play_random_game(struct playout_setup
*setup
,
102 struct board
*b
, enum stone starting_color
,
103 struct playout_amafmap
*amafmap
,
104 struct board_ownermap
*ownermap
,
105 struct playout_policy
*policy
);