11import json
2- from typing import Any , Dict , NoReturn
2+ from typing import Any , Tuple , List , Dict , NoReturn
33
44from program import PROGRAM_NAME , VERSION , PROTOCOL_VERSION
5- from board .go_board import GoBoard
5+ from board .go_board import GoBoard , copy_board
66from board .coordinate import Coordinate
77from board .stone import Stone
88from mcts .constant import NOT_EXPANDED
@@ -19,10 +19,12 @@ def dump_mcts_to_json(tree_dict: Dict[str, Any], board: GoBoard, superko: bool)
1919 str: MCTSの状態を表すJSON文字列。
2020 """
2121 state = {
22- "dump_version" : 1 ,
22+ "dump_version" : 2 ,
2323 "tree" : tree_dict ,
2424 "board_size" : board .get_board_size (),
2525 "komi" : board .get_komi (),
26+ "move_history" : _serializable_move_history (board .get_move_history ()),
27+ "handicap_history" : board .get_handicap_history (),
2628 "superko" : superko ,
2729 "name" : PROGRAM_NAME ,
2830 "version" : VERSION ,
@@ -36,7 +38,12 @@ def enrich_mcts_dict(state: Dict[str, Any]) -> NoReturn:
3638 Args:
3739 state (Dict[str, Any]): MCTSの状態を表す辞書。
3840 """
39- coord = Coordinate (board_size = state ["board_size" ])
41+ root_board = GoBoard (board_size = state ["board_size" ], komi = state ["komi" ], \
42+ check_superko = state ["superko" ])
43+ root_board .set_history (_recovered_move_history (state ["move_history" ]), \
44+ state ["handicap_history" ])
45+
46+ coord = Coordinate (board_size = root_board .get_board_size ())
4047 tree = state ["tree" ]
4148 node = tree ["node" ]
4249
@@ -73,24 +80,31 @@ def enrich_mcts_dict(state: Dict[str, Any]) -> NoReturn:
7380 nodes_pool += expanded_children
7481
7582 # その他いろいろな便利項目を追加
83+ initial_move_color = _str_to_stone (tree ["to_move" ])
7684 for item in node :
7785 is_root = "parent_index" not in item
7886 if is_root :
7987 item ["level" ] = 0
8088 item ["orders_along_path" ] = []
89+ item ["gtp_moves_along_path" ] = []
8190 item ["to_move" ] = tree ["to_move" ]
91+ item ["board_string" ] = root_board .get_board_string ()
8292 continue
8393 parent = node [item ["parent_index" ]]
94+ index_in_brother = item ["index_in_brother" ]
95+ gtp_move = coord .convert_to_gtp_format (parent ["action" ][index_in_brother ])
8496 item ["level" ] = parent ["level" ] + 1
8597 item ["orders_along_path" ] = [* parent ["orders_along_path" ], item ["order" ]]
8698 item ["to_move" ] = _opposite_color (parent ["to_move" ])
99+ item ["gtp_moves_along_path" ] = [* parent ["gtp_moves_along_path" ], gtp_move ]
100+ item ["board_string" ] = _get_updated_board_string (root_board , initial_move_color , \
101+ item ["gtp_moves_along_path" ])
87102 # ルートノードは以下の項目を持たないことに注意
88- index_in_brother = item ["index_in_brother" ]
89103 item ["policy" ] = parent ["children_policy" ][index_in_brother ]
90104 item ["visits" ] = parent ["children_visits" ][index_in_brother ]
91105 item ["value" ] = parent ["children_value" ][index_in_brother ]
92106 item ["value_sum" ] = parent ["children_value_sum" ][index_in_brother ]
93- item ["gtp_move" ] = coord . convert_to_gtp_format ( parent [ "action" ][ index_in_brother ])
107+ item ["gtp_move" ] = gtp_move
94108 item ["mean_value" ] = item ["value_sum" ] / item ["visits" ]
95109 last_move_color = _opposite_color (item ["to_move" ])
96110 item ["raw_black_winrate" ] = _black_winrate (item ["value" ], last_move_color )
@@ -101,3 +115,54 @@ def _opposite_color(color):
101115
102116def _black_winrate (value , last_move_color ):
103117 return value if last_move_color == "black" else 1.0 - value
118+
119+ def _serializable_move_history (move_history : List [Tuple [Stone , int , Any ]]) -> List [Tuple [str , int ]]:
120+ """着手の履歴をシリアライズ可能な値に変換する。ただしハッシュ値は廃棄する。
121+
122+ Args:
123+ move_history (List[Tuple[Stone, int, np.array]]): 着手の履歴。
124+
125+ Returns:
126+ Lizt[Tuple[str, int]]: シリアライズ可能なよう変換された着手履歴。
127+ """
128+ return [(_stone_to_str (color ), pos ) for (color , pos , _ ) in move_history ]
129+
130+ def _recovered_move_history (converted_move_history : List [Tuple [str , int ]]) -> List [Tuple [Stone , int , Any ]]:
131+ """_serializable_move_historyで変換された着手履歴から元の着手履歴を復元する。
132+ ただしハッシュ値はNoneに置きかえられる。
133+
134+ Args:
135+ converted_move_history (Lizt[Tuple[str, int]]): 変換された着手履歴。
136+
137+ Returns:
138+ List[Tuple[Stone, int, Any]]: 復元された着手履歴。
139+ """
140+ return [(_str_to_stone (color_str ), pos , None ) for (color_str , pos ) in converted_move_history ]
141+
142+ def _stone_to_str (color : Stone ) -> str :
143+ return 'black' if color == Stone .BLACK else 'white'
144+
145+ def _str_to_stone (color_str : str ) -> str :
146+ return Stone .BLACK if color_str == 'black' else Stone .WHITE
147+
148+ def _get_updated_board_string (root_board : GoBoard , initial_move_color : Stone , gtp_moves_along_path : List [str ]) -> str :
149+ """一連の着手後の盤面を表わす文字列を返す。
150+
151+ Args:
152+ root_board (GoBoard): 着手前の盤面。
153+ initial_move_color (Stone): 最初の着手の色。
154+ gtp_moves_along_path (List[str]): 着手位置のリスト。
155+
156+ Returns:
157+ str: 着手後の盤面を表わす文字列。
158+ """
159+ coord = Coordinate (board_size = root_board .get_board_size ())
160+ move_color = initial_move_color
161+ # 「board = copy.deepcopy(root_board)」は遅いので避ける。
162+ board = GoBoard (board_size = root_board .get_board_size (), komi = root_board .get_komi (), check_superko = root_board .check_superko )
163+ copy_board (dst = board , src = root_board )
164+ for (k , move ) in enumerate (gtp_moves_along_path ):
165+ pos = coord .convert_from_gtp_format (move )
166+ board .put_stone (pos , move_color )
167+ move_color = Stone .get_opponent_color (move_color )
168+ return board .get_board_string ()
0 commit comments