rave_mctsagent.py

class DecisiveMoveMctsAgent(RaveMctsAgent):
    def roll_out(self, state: GameState) -> tuple:
        """
        Simulate a random game except that we play all known critical cells
        first, return the winning player and record critical cells at the end.
        """
        moves = state.moves()
        good_moves = moves.copy()
        good_opponent_moves = moves.copy()
        to_play = state.turn()

        while state.winner == GameMeta.PLAYERS["none"]:
            done = False
            # In each state transition, check if there is any critical move between all possible moves.
            while len(good_moves) > 0 and not done:
                move = choice(good_moves)
                good_moves.remove(move)
                if not state.would_lose(move, to_play):
                    # is this chosen move ending in win
                    state.play(move)
                    moves.remove(move)
                    if move in good_opponent_moves:
                        good_opponent_moves.remove(move)
                    done = True

            if not done:
                move = choice(moves)
                state.play(move)
                moves.remove(move)
                if move in good_opponent_moves:
                    good_opponent_moves.remove(move)
            # state has changed, so good moves are exchanged between players.
            good_moves, good_opponent_moves = good_opponent_moves, good_moves

        black_rave_pts = []
        white_rave_pts = []

        for x in range(state.size):
            for y in range(state.size):
                if state.board[(x, y)] == GameMeta.PLAYERS["black"]:
                    black_rave_pts.append((x, y))
                elif state.board[(x, y)] == GameMeta.PLAYERS["white"]:
                    white_rave_pts.append((x, y))

        return state.winner, black_rave_pts, white_rave_pts