DDR爱好者之家 Design By 杰米

利用上一篇的框架,再写了个翻转棋的程序,为了调试minimax算法,花了两天的时间。

几点改进说明:

  • 拆分成四个文件:board.py,player.py,ai.py,othello.py。使得整个结构更清晰,更通用,更易于维护。
  • AI 的水平跟 minimax 的递归深度,以及评价函数有关。基于此,我把 minimax 和评价函数都放到 AI 类里面
  • AIPlayer 使用了多重继承。继承了 Player 与 AI 两个类
  • Game 类中把原run函数里的生成两个玩家的部分提出来,写成一个函数make_two_players,使得 run函数结构更清晰
  • AI 玩家等级不要选择 0:beginer。会报错,还没调试好

board.py

'''
作者:hhh5460
时间:2017年7月1日
'''

class Board(object):
 def __init__(self):
  self.empty = '.'
  self._board = [[self.empty for _ in range(8)] for _ in range(8)] # 规格:8*8
  self._board[3][4], self._board[4][3] = 'X', 'X'
  self._board[3][3], self._board[4][4] = 'O', 'O'
  
 # 增加 Board[][] 索引语法
 def __getitem__(self, index):
  return self._board[index]
 
 # 打印棋盘
 def print_b(self):
  board = self._board
  print(' ', ' '.join(list('ABCDEFGH')))
  for i in range(8):
   print(str(i+1),' '.join(board[i]))
   
 # 棋局终止
 def teminate(self):
  list1 = list(self.get_legal_actions('X'))
  list2 = list(self.get_legal_actions('O'))
  return [False, True][len(list1) == 0 and len(list2) == 0]
  
 # 判断赢家
 def get_winner(self):
  s1, s2 = 0, 0
  for i in range(8):
   for j in range(8):
    if self._board[i][j] == 'X':
     s1 += 1
    if self._board[i][j] == 'O':
     s2 += 1
  if s1 > s2:
   return 0 # 黑胜
  elif s1 < s2:
   return 1 # 白胜
  elif s1 == s2:
   return 2 # 平局
 # 落子
 def _move(self, action, color):
  x,y = action
  self._board[x][y] = color
  
  return self._flip(action, color)
  
 
  
 
 # 翻子(返回list)
 def _flip(self, action, color):
  flipped_pos = []
  
  for line in self._get_lines(action):
   for i,p in enumerate(line):
    if self._board[p[0]][p[1]] == self.empty: 
     break
    elif self._board[p[0]][p[1]] == color:
     flipped_pos.extend(line[:i])
     break
  
  for p in flipped_pos:
   self._board[p[0]][p[1]] = color
   
  return flipped_pos
  
 # 撤销
 def _unmove(self, action, flipped_pos, color):
  self._board[action[0]][action[1]] = self.empty
  
  uncolor = ['X', 'O'][color=='X']
  for p in flipped_pos:
   self._board[p[0]][p[1]] = uncolor
   
 # 生成8个方向的下标数组,方便后续操作
 def _get_lines(self, action):
  '''说明:刚开始我是用一维棋盘来考虑的,后来改为二维棋盘。偷懒,不想推倒重来,简单地修改了一下'''
  board_coord = [(i,j) for i in range(8) for j in range(8)] # 棋盘坐标
  
  r,c = action
  ix = r*8 + c
  r, c = ix//8, ix%8
  left = board_coord[r*8:ix] # 要反转
  right = board_coord[ix+1:(r+1)*8]
  top = board_coord[c:ix:8] # 要反转
  bottom = board_coord[ix+8:8*8:8]
  
  if r <= c:
   lefttop = board_coord[c-r:ix:9] # 要反转
   rightbottom = board_coord[ix+9:(7-(c-r))*8+7+1:9]
  else:
   lefttop = board_coord[(r-c)*8:ix:9] # 要反转
   rightbottom = board_coord[ix+9:7*8+(7-(c-r))+1:9]
  
  if r+c<=7:
   leftbottom = board_coord[ix+7:(r+c)*8:7]
   righttop = board_coord[r+c:ix:7] # 要反转
  else:
   leftbottom = board_coord[ix+7:7*8+(r+c)-7+1:7]
   righttop = board_coord[((r+c)-7)*8+7:ix:7] # 要反转
  
  # 有四个要反转,方便判断
  left.reverse()
  top.reverse()
  lefttop.reverse()
  righttop.reverse()
  lines = [left, top, lefttop, righttop, right, bottom, leftbottom, rightbottom]
  return lines
  
 # 检测,位置是否有子可翻
 def _can_fliped(self, action, color):
  flipped_pos = []
  
  for line in self._get_lines(action):
   for i,p in enumerate(line):
    if self._board[p[0]][p[1]] == self.empty: 
     break
    elif self._board[p[0]][p[1]] == color:
     flipped_pos.extend(line[:i])
     break
  return [False, True][len(flipped_pos) > 0]
  
 # 合法走法
 def get_legal_actions(self, color):
  uncolor = ['X', 'O'][color=='X']
  uncolor_near_points = [] # 反色邻近的空位
  
  board = self._board
  for i in range(8):
   for j in range(8):
    if board[i][j] == uncolor:
     for dx,dy in [(-1,0),(-1,1),(0,1),(1,1),(1,0),(1,-1),(0,-1)]:
      x, y = i+dx, j+dy
      if 0 <= x <=7 and 0 <= y <=7 and board[x][y] == self.empty and (x, y) not in uncolor_near_points:
       uncolor_near_points.append((x, y))
  for p in uncolor_near_points:
   if self._can_fliped(p, color):
    yield p

# 测试
if __name__ == '__main__':
 board = Board()
 board.print_b()
 print(list(board.get_legal_actions('X')))

player.py

from ai import AI

'''
作者:hhh5460
时间:2017年7月1日
'''

# 玩家
class Player(object):
 def __init__(self, color):
  self.color = color
  
 # 思考
 def think(self, board):
  pass
  
 # 落子
 def move(self, board, action):
  flipped_pos = board._move(action, self.color)
  return flipped_pos
  
 # 悔子
 def unmove(self, board, action, flipped_pos):
  board._unmove(action, flipped_pos, self.color)


# 人类玩家
class HumanPlayer(Player):
 def __init__(self, color):
  super().__init__(color)
 
 def think(self, board):
  while True:
   action = input("Turn to '{}'. \nPlease input a point.(such as 'A1'): ".format(self.color)) # A1~H8
   r, c = action[1], action[0].upper()
   if r in '12345678' and c in 'ABCDEFGH': # 合法性检查1
    x, y = '12345678'.index(r), 'ABCDEFGH'.index(c)
    if (x,y) in board.get_legal_actions(self.color): # 合法性检查2
     return x, y


# 电脑玩家(多重继承)
class AIPlayer(Player, AI):
 
 def __init__(self, color, level_ix=0):
  super().__init__(color)    # init Player
  super(Player, self).__init__(level_ix) # init AI
  
 
 def think(self, board):
  print("Turn to '{}'. \nPlease wait a moment. AI is thinking...".format(self.color))
  uncolor = ['X','O'][self.color=='X']
  opfor = AIPlayer(uncolor) # 假想敌,陪练
  action = self.brain(board, opfor, 4)
  return action

ai.py

import random

'''
作者:hhh5460
时间:2017年7月1日
'''

class AI(object):
 '''
 三个水平等级:初级(beginner)、中级(intermediate)、高级(advanced)
 '''
 def __init__(self, level_ix =0):
  # 玩家等级
  self.level = ['beginner','intermediate','advanced'][level_ix]
  # 棋盘位置权重,参考:https://github.com/k-time/ai-minimax-agent/blob/master/ksx2101.py
  self.board_weights = [
   [120, -20, 20, 5, 5, 20, -20, 120],
   [-20, -40, -5, -5, -5, -5, -40, -20],
   [ 20, -5, 15, 3, 3, 15, -5, 20],
   [ 5, -5, 3, 3, 3, 3, -5, 5],
   [ 5, -5, 3, 3, 3, 3, -5, 5],
   [ 20, -5, 15, 3, 3, 15, -5, 20],
   [-20, -40, -5, -5, -5, -5, -40, -20],
   [120, -20, 20, 5, 5, 20, -20, 120]
  ]
  
 # 评估函数(仅根据棋盘位置权重)
 def evaluate(self, board, color):
  uncolor = ['X','O'][color=='X']
  score = 0
  for i in range(8):
   for j in range(8):
    if board[i][j] == color:
     score += self.board_weights[i][j]
    elif board[i][j] == uncolor:
     score -= self.board_weights[i][j]
  return score

 # AI的大脑
 def brain(self, board, opponent, depth):
  if self.level == 'beginer':   # 初级水平
   _, action = self.randomchoice(board)
  elif self.level == 'intermediate': # 中级水平
   _, action = self.minimax(board, opponent, depth)
  elif self.level == 'advanced':  # 高级水平
   _, action = self.minimax_alpha_beta(board, opponent, depth)
  assert action is not None, 'action is None'
  return action
 
 # 随机选(从合法走法列表中随机选)
 def randomchoice(self, board):
  color = self.color
  action_list = list(board.get_legal_actions(color))
  return None, random.choice(action_list)
 
 # 极大极小算法,限制深度
 def minimax(self, board, opfor, depth=4): # 其中 opfor 是假想敌、陪练
  '''参考:https://github.com/k-time/ai-minimax-agent/blob/master/ksx2101.py'''
  color = self.color
  
  if depth == 0:
   return self.evaluate(board, color), None
  
  action_list = list(board.get_legal_actions(color))
  if not action_list:
   return self.evaluate(board, color), None
  
  best_score = -100000
  best_action = None

  for action in action_list:
   flipped_pos = self.move(board, action) # 落子
   score, _ = opfor.minimax(board, self, depth-1) # 深度优先,轮到陪练
   self.unmove(board, action, flipped_pos) # 回溯
   
   score = -score
   if score > best_score:
    best_score = score
    best_action = action

  return best_score, best_action
  
 # 极大极小算法,带alpha-beta剪枝
 def minimax_alpha_beta(self, board, opfor, depth=8, my_best=-float('inf'), opp_best=float('inf')):
  '''参考:https://github.com/k-time/ai-minimax-agent/blob/master/ksx2101.py'''
  color = self.color
  
  if depth == 0:
   return self.evaluate(board, color), None
  
  action_list = list(board.get_legal_actions(color))
  if not action_list:
   return self.evaluate(board, color), None
  
  best_score = my_best
  best_action = None
  
  for action in action_list:
   flipped_pos = self.move(board, action) # 落子
   score, _ = opfor.minimax_alpha_beta(board, self, depth-1, -opp_best, -best_score) # 深度优先,轮到陪练
   self.unmove(board, action, flipped_pos) # 回溯
   
   score = -score
   if score > best_score:
    best_score = score
    best_action = action
    
   if best_score > opp_best:
    break

  return best_score, best_action

othello.py

from board import Board
from player import HumanPlayer, AIPlayer

'''
作者:hhh5460
时间:2017年7月1日
'''

# 游戏
class Game(object):
 def __init__(self):
  self.board = Board()
  self.current_player = None
  
 # 生成两个玩家
 def make_two_players(self):
  ps = input("Please select two player's type:\n\t0.Human\n\t1.AI\nSuch as:0 0\n:")
  p1, p2 = [int(p) for p in ps.split(' ')]
  if p1 == 1 or p2 == 1: # 至少有一个AI玩家
   level_ix = int(input("Please select the level of AI player.\n\t0: beginner\n\t1: intermediate\n\t2: advanced\n:"))
   if p1 == 0:
    player1 = HumanPlayer('X')
    player2 = AIPlayer('O', level_ix)
   elif p2 == 0:
    player1 = AIPlayer('X', level_ix)
    player2 = HumanPlayer('O')
   else:
    player1 = AIPlayer('X', level_ix)
    player2 = AIPlayer('O', level_ix)
  else:
   player1, player2 = HumanPlayer('X'), HumanPlayer('O') # 先手执X,后手执O
  
  return player1, player2
 
 
 # 切换玩家(游戏过程中)
 def switch_player(self, player1, player2):
  if self.current_player is None:
   return player1
  else:
   return [player1, player2][self.current_player == player1]
 
 # 打印赢家
 def print_winner(self, winner): # winner in [0,1,2]
  print(['Winner is player1','Winner is player2','Draw'][winner])
 
 # 运行游戏
 def run(self):
  # 生成两个玩家
  player1, player2 = self.make_two_players()
  
  # 游戏开始
  print('\nGame start!\n')
  self.board.print_b() # 显示棋盘
  while True:
   self.current_player = self.switch_player(player1, player2) # 切换当前玩家
   
   action = self.current_player.think(self.board) # 当前玩家对棋盘进行思考后,得到招法
   
   if action is not None: 
    self.current_player.move(self.board, action) # 当前玩家执行招法,改变棋盘
   
   self.board.print_b() # 显示当前棋盘
   
   if self.board.teminate(): # 根据当前棋盘,判断棋局是否终止
    winner = self.board.get_winner() # 得到赢家 0,1,2
    break
  
  self.print_winner(winner)
  print('Game over!')
  
  self.board.print_history()
 
 
if __name__ == '__main__':
 Game().run()

效果图

python实现翻转棋游戏(othello)

以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持。

DDR爱好者之家 Design By 杰米
广告合作:本站广告合作请联系QQ:858582 申请时备注:广告合作(否则不回)
免责声明:本站资源来自互联网收集,仅供用于学习和交流,请遵循相关法律法规,本站一切资源不代表本站立场,如有侵权、后门、不妥请联系本站删除!
DDR爱好者之家 Design By 杰米

稳了!魔兽国服回归的3条重磅消息!官宣时间再确认!

昨天有一位朋友在大神群里分享,自己亚服账号被封号之后居然弹出了国服的封号信息对话框。

这里面让他访问的是一个国服的战网网址,com.cn和后面的zh都非常明白地表明这就是国服战网。

而他在复制这个网址并且进行登录之后,确实是网易的网址,也就是我们熟悉的停服之后国服发布的暴雪游戏产品运营到期开放退款的说明。这是一件比较奇怪的事情,因为以前都没有出现这样的情况,现在突然提示跳转到国服战网的网址,是不是说明了简体中文客户端已经开始进行更新了呢?