Fixing a little problem in playouts

game/strategy.py

@@ -155,7 +155,7 @@ class StrategyAlphaBeta(Strategy):
 
     def get_first_move(self) -> Action:
         """
-        This function returns the first move for the Gopher game, calculated by the precompute_firs_move
+        This function returns the first move for the Gopher game, calculated by the precompute_first_move
         It is meant to save the time wasted by the alphabeta when calculating the very first move of the game, which
         takes a very long time because there are a lot of possibilities when the board is empty.
         Therefore, the first move is calculated beforehand
@@ -168,12 +168,12 @@ class StrategyAlphaBeta(Strategy):
     def get_action(self, grid: HexGrid, ply: Player) -> Action:
         self.current_state = grid
 
-        action = None
-        if self.rules.is_first_turn(grid) and isinstance(self.rules, Gopher):
-            action = self.get_first_move()
+        # action = None
+        # if self.rules.is_first_turn(grid) and isinstance(self.rules, Gopher):
+        #    action = self.get_first_move()
 
-        if action is not None:
-            return action
+        # if action is not None:
+        #    return action
         return self.minmax(ply, -inf, inf, self.depth)[1]
 
 
@@ -313,7 +313,7 @@ class MCTSStrategy(Strategy):
                 if res is not None:
                     action = res[1]
 
-            if action is None:
+            if action is None or action not in legals:
                 i = self.rng.choice(range(len(legals)), 1)[0]
                 action = legals[i]
 
@@ -493,16 +493,6 @@ class MCTSStrategy(Strategy):
         """
         Performs a training of the MCTS tree.
         """
-
-        # We need to re-init the see for the
-        # random number generators so that
-        # it's not always the same nodes expanded
-        # for each simulation accross the different
-        # processes
-        seed = random.SeedSequence(int.from_bytes(os.urandom(16), "big") * os.getpid())
-        rng = random.Generator(random.MT19937(seed))
-        self.rng = rng
-
         root = self.root
         node = self.__select(root)
         self.__expand(node, self.rules)
@@ -518,6 +508,16 @@ class MCTSStrategy(Strategy):
         Args:
             simulations: The number of simulations to run
         """
+
+        # We need to re-init the see for the
+        # random number generators so that
+        # it's not always the same nodes expanded
+        # for each simulation accross the different
+        # processes
+        seed = random.SeedSequence(int.from_bytes(os.urandom(16), "big") * os.getpid())
+        rng = random.Generator(random.MT19937(seed))
+        self.rng = rng
+
         for _ in range(simulations):
             self.__train()
         return self.root