simulator process a valid set, instead of a single action

AlphaGo/go.py

@@ -121,8 +121,8 @@ class Go:
         if self._is_eye(current_board, color, vertex):
             return False
             # forbid position on its own eye.
-        if self._is_game_finish(current_board, color) and vertex == utils.PASS
-            return False
+        #if self._is_game_finish(current_board, color) and vertex == utils.PASS
+        #    return False
         # forbid pass if the game is not finished.
         return True
 
@@ -183,6 +183,18 @@ class Go:
 
         return True
 
+    def simulate_is_valid_list(self, state, action_set):
+        ## find all the valid actions
+        ## if no action is valid, then pass
+        valid_action_set = []
+        for action_candidate in action_set:
+            if self.simulate_is_valid(self, state, action_candidate)
+                valid_action_set.append(action_candidate)
+        if not valid_action_set:
+            valid_action_set.append(utils.PASS)
+            # if valid_action_set is a empty set, add pass
+        return valid_action_set
+
     def _do_move(self, board, color, vertex):
         if vertex == utils.PASS:
             return board

tianshou/core/mcts/mcts.py

@@ -72,12 +72,7 @@ class UCTNode(MCTSNode):
 
     def valid_mask(self, simulator):
         if self.mask is None:
-            self.mask = []
-            for act in range(self.action_num - 1):
-                if not simulator.simulate_is_valid(self.state, act):
-                    self.mask.append(act)
-                    self.ucb[act] = -float("Inf")
-        else:
+            self.mask = simulator.simulate_is_valid_list(self.state, range(self.action_num - 1))
         self.ucb[self.mask] = -float("Inf")