| |
| |
| @@ -40,25 +40,31 @@ |
| |
| def step(self, action: Connect4Action): |
| col = action.column |
| - # reward = 0.0 |
| done = False |
| + reward = 0.0 |
| |
| # check action validity |
| if col < 0 or col >= self.COLUMNS or self.board[0, col] != 0: |
| self.invalid_move_played = True |
| - reward = -1 # penalty for invalid move |
| + reward = -1.0 # penalty for invalid move |
| done = True |
| else: |
| - # drop piece |
| - for row in range(self.ROWS - 1, -1, -1): |
| - if self.board[row, col] == 0: |
| - self.board[row, col] = self.next_player |
| - break |
| - |
| - # check win / full board |
| - reward, done = self._check_win_or_draw(row, col) |
| - |
| - self.next_player *= -1 |
| + # Agent is always player 1. |
| + agent_row = self._drop_piece(col, 1) |
| + reward, done = self._check_win_or_draw(agent_row, col) |
| + |
| + # Let the opponent respond when the game is still in progress. |
| + if not done: |
| + opponent_col = self._choose_opponent_move() |
| + if opponent_col is None: |
| + reward = 0.0 |
| + done = True |
| + else: |
| + opponent_row = self._drop_piece(opponent_col, -1) |
| + reward, done = self._check_win_or_draw(opponent_row, opponent_col) |
| + |
| + # The agent moves first again on the next turn. |
| + self.next_player = 1 |
| |
| self._state = Connect4State( |
| board=self.board.copy().tolist(), |
| @@ -79,27 +85,84 @@ |
| metadata={"next_player": self.next_player}, |
| ) |
| |
| - def _check_win_or_draw(self, row, col): |
| - # Implement 4-in-a-row check (like your Gymnasium code) |
| + def _drop_piece(self, col, player): |
| + for row in range(self.ROWS - 1, -1, -1): |
| + if self.board[row, col] == 0: |
| + self.board[row, col] = player |
| + return row |
| + return -1 |
| + |
| + def _find_drop_row(self, col): |
| + for row in range(self.ROWS - 1, -1, -1): |
| + if self.board[row, col] == 0: |
| + return row |
| + return -1 |
| + |
| + def _is_four(self, row, col): |
| player = self.board[row, col] |
| directions = [(1, 0), (0, 1), (1, 1), (1, -1)] |
| for dr, dc in directions: |
| - count = 0 |
| - for step in range(-3, 4): |
| - r, c = row + step * dr, col + step * dc |
| - if ( |
| - 0 <= r < self.ROWS |
| - and 0 <= c < self.COLUMNS |
| - and self.board[r, c] == player |
| - ): |
| - count += 1 |
| - if count >= 4: |
| - return 1.0, True |
| - else: |
| - count = 0 |
| + count = 1 |
| + for step in (-1, 1): |
| + r, c = row, col |
| + while True: |
| + r += step * dr |
| + c += step * dc |
| + if ( |
| + 0 <= r < self.ROWS |
| + and 0 <= c < self.COLUMNS |
| + and self.board[r, c] == player |
| + ): |
| + count += 1 |
| + else: |
| + break |
| + if count >= 4: |
| + return True |
| + return False |
| + |
| + def _check_win_or_draw(self, row, col): |
| + player = self.board[row, col] |
| + if self._is_four(row, col): |
| + # +1 when the agent (player 1) wins, -1 when the opponent wins. |
| + return (1.0 if player == 1 else -1.0), True |
| if np.all(self.board != 0): |
| return 0.0, True |
| return 0.0, False |
| + |
| + def _choose_opponent_move(self): |
| + legal_actions = [c for c in range(self.COLUMNS) if self.board[0, c] == 0] |
| + if not legal_actions: |
| + return None |
| + |
| + # Allow callers to inject a custom opponent strategy. |
| + if callable(self._opponent): |
| + try: |
| + col = self._opponent(self.board.copy(), legal_actions) |
| + except TypeError: |
| + col = self._opponent(legal_actions) |
| + if col in legal_actions: |
| + return col |
| + |
| + # Prefer an immediate winning move. |
| + for col in legal_actions: |
| + row = self._find_drop_row(col) |
| + self.board[row, col] = -1 |
| + wins = self._is_four(row, col) |
| + self.board[row, col] = 0 |
| + if wins: |
| + return col |
| + |
| + # Otherwise block an immediate agent win. |
| + for col in legal_actions: |
| + row = self._find_drop_row(col) |
| + self.board[row, col] = 1 |
| + wins = self._is_four(row, col) |
| + self.board[row, col] = 0 |
| + if wins: |
| + return col |
| + |
| + # Fall back to the most central available column. |
| + return min(legal_actions, key=lambda c: (abs(c - 3), c)) |
| |
| def web_actions( |
| self, observation: Dict[str, Any] |
|
|
