随机切换明牌模式进行训练

douzero/env/env.py

@@ -130,6 +130,7 @@ class Env:
         # Initialize the internal environment
         self._env = GameEnv(self.players)
         self.total_round = 0
+        self.face_up_level = 0
         self.infoset = None
 
     def reset(self, flags=None):
@@ -153,6 +154,13 @@ class Env:
                           }
         for key in card_play_data:
             card_play_data[key].sort()
+        rint = np.random.randint(0, 100)
+        if rint < 45:
+            face_up_level = 0
+        elif rint < 90:
+            face_up_level = 0x01
+        else:
+            face_up_level = 0x02
         player_ids = {
             'landlord': 0,
             'landlord_down': 1,
@@ -165,6 +173,7 @@ class Env:
         for pos in ["landlord", "landlord_up", "landlord_front", "landlord_down"]:
             pid = player_ids[pos]
             self._env.info_sets[pos].player_id = pid
+            self._env.info_sets[pos].face_up_level = face_up_level
         self.infoset = self._game_infoset
 
         return get_obs(self.infoset, self.use_general, self.use_legacy, self.lite_model, self.use_unified)
@@ -1065,10 +1074,7 @@ def _get_obs_general(infoset, position, compressed_form = False):
     }
     return obs
 
-'''
+def _get_obs_unified(infoset, position, compressed_form = True):
-    face_up_level   0x01: three_landlord_cards, 0x02: landlord, 0x04: landlord_up, 0x08: landlord_front, 0x10: landlord_down
-'''
-def _get_obs_unified(infoset, position, compressed_form = True, face_up_level = 0):
     num_legal_actions = len(infoset.legal_actions)
     my_handcards = _cards2array(infoset.player_hand_cards, compressed_form)
     my_handcards_batch = np.repeat(my_handcards[np.newaxis, :],
@@ -1104,7 +1110,7 @@ def _get_obs_unified(infoset, position, compressed_form = True, face_up_level =
     landlord_down_played_cards = _cards2array(
         infoset.played_cards['landlord_down'], compressed_form)
 
-    if (face_up_level & 0x01) > 0:
+    if (infoset.face_up_level & 0x01) > 0:
         three_landlord_cards = _cards2array(
             infoset.three_landlord_cards, compressed_form)
 
@@ -1117,28 +1123,28 @@ def _get_obs_unified(infoset, position, compressed_form = True, face_up_level =
         three_landlord_cards_all = _cards2noise(
             infoset.three_landlord_cards_all, compressed_form)
 
-    if (face_up_level & 0x02) > 0:
+    if (infoset.face_up_level & 0x02) > 0:
         landlord_cards = _cards2array(
             infoset.all_handcards['landlord'], compressed_form)
     else:
         landlord_cards = _cards2noise(
             infoset.all_handcards['landlord'], compressed_form)
 
-    if (face_up_level & 0x04) > 0:
+    if (infoset.face_up_level & 0x04) > 0:
         landlord_up_cards = _cards2array(
             infoset.all_handcards['landlord_up'], compressed_form)
     else:
         landlord_up_cards = _cards2noise(
             infoset.all_handcards['landlord_up'], compressed_form)
 
-    if (face_up_level & 0x08) > 0:
+    if (infoset.face_up_level & 0x08) > 0:
         landlord_front_cards = _cards2array(
             infoset.all_handcards['landlord_front'], compressed_form)
     else:
         landlord_front_cards = _cards2noise(
             infoset.all_handcards['landlord_front'], compressed_form)
 
-    if (face_up_level & 0x10) > 0:
+    if (infoset.face_up_level & 0x10) > 0:
         landlord_down_cards = _cards2array(
             infoset.all_handcards['landlord_down'], compressed_form)
     else:
@@ -1149,7 +1155,7 @@ def _get_obs_unified(infoset, position, compressed_form = True, face_up_level =
         infoset.bomb_num, compressed_form=compressed_form) # 56/95
     base_info = np.hstack((
         PositionInfoArray[position], # 4
-        FaceUpLevelArray[face_up_level], # 9
+        FaceUpLevelArray[infoset.face_up_level], # 9
         bomb_num, #56
     ))
     num_cards_left = np.hstack((

douzero/env/game.py

@@ -168,6 +168,7 @@ class GameEnv(object):
             "landlord_down": 0
         }
         self.last_pid = 'landlord'
+        self.face_up_level = 0
         self.step_count = 0
 
 
@@ -192,6 +193,16 @@ class GameEnv(object):
         else:
             self.three_landlord_cards = card_play_data['three_landlord_cards'][:]
             self.three_landlord_cards_all = card_play_data['three_landlord_cards'][:]
+        if 'face_up_level' in card_play_data.keys():
+            self.info_sets['landlord'].face_up_level = card_play_data['face_up_level']
+            self.info_sets['landlord_up'].face_up_level = card_play_data['face_up_level']
+            self.info_sets['landlord_front'].face_up_level = card_play_data['face_up_level']
+            self.info_sets['landlord_down'].face_up_level = card_play_data['face_up_level']
+        else:
+            self.info_sets['landlord'].face_up_level = 0
+            self.info_sets['landlord_up'].face_up_level = 0
+            self.info_sets['landlord_front'].face_up_level = 0
+            self.info_sets['landlord_down'].face_up_level = 0
         self.get_acting_player_position()
         self.game_infoset = self.get_infoset()
 
@@ -364,6 +375,11 @@ class GameEnv(object):
                          'landlord_front': InfoSet('landlord_front'),
                          'landlord_down': InfoSet('landlord_down')}
 
+        self.info_sets['landlord'].face_up_level = self.face_up_level
+        self.info_sets['landlord_up'].face_up_level = self.face_up_level
+        self.info_sets['landlord_front'].face_up_level = self.face_up_level
+        self.info_sets['landlord_down'].face_up_level = self.face_up_level
+
         self.bomb_num = [0, 0, 0]
         self.pos_bomb_num = {
             "landlord": 0,
@@ -456,3 +472,5 @@ class InfoSet(object):
         # The number of bombs played so far
         self.bomb_num = None
         self.player_id = None
+        # face_up_level   0x01: three_landlord_cards, 0x02: landlord, 0x04: landlord_up, 0x08: landlord_front, 0x10: landlord_down
+        self.face_up_level = 0