AI-wuziqi游戏.zip

from .rules import * import random import json import math attempts = 25 totAttempts = 4000 curAttempts = 0 record = {} def Load(fileId): fileName = "./wuziqi/record/file%d.json" % fileId with open(fileName, 'r') as f: record=json.load(f) def Dump(fileId): fileName = "./wuziqi/record/file%d.json" % fileId with open(fileName, 'w') as f: json.dump(record,f) def CurStep(board): res = 0 for i in range(board.shape[0]): for j in range(board.shape[1]): res += 1 if board[i,j] != 0 else 0 return res def Serialize(board): res = '' for i in range(board.shape[0]): tmp = 0 for j in range(board.shape[1]): tmp = tmp * 3 + board[i,j] res += str(tmp) return res def GetLegalPositions(board): res = [] for i in range(board.shape[0]): for j in range(board.shape[1]): if board[i,j] == EMPTY: res.append((i,j)) return res def RandomStep(legal, value): item = np.random.choice(range(value.shape[0]), 1, True, value)[0] return legal[item] def ValueFunc(board, player, pos): x, y = pos value = 0 if pos[0] == 0 or pos[0] == board.shape[0] or pos[1] == 0 or pos[1] == board.shape[1]: value -= 1 # 先找自己能挡多少 for d in directions: cnt = 0 cx, cy = x, y while True: cx += d[0] cy += d[1] if(PositionInBoard(board, cx, cy) and board[cx, cy] == 3-player): cnt += 1 else: break cx, cy = x, y while True: cx -= d[0] cy -= d[1] if(PositionInBoard(board, cx, cy) and board[cx, cy] == 3-player): cnt += 1 else: break if (cnt >= 3): value += 20 elif (cnt >= 2): value += 8 elif (cnt >= 1): value += 2 # 再找自己能连多少 numberOf3 = 0 for d in directions: cnt = 1 cx, cy = x, y while True: cx += d[0] cy += d[1] if(PositionInBoard(board, cx, cy) and board[cx, cy] == player): cnt += 1 else: break cx, cy = x, y while True: cx -= d[0] cy -= d[1] if(PositionInBoard(board, cx, cy) and board[cx, cy] == player): cnt += 1 else: break if (cnt >= 5): value += 100 if (cnt >= 4): value += 20 elif (cnt >= 3): value += 8 elif (cnt >= 2): value += 4 elif (cnt >= 1): value += 2 if (cnt >= 3): numberOf3 += 1 if numberOf3 >= 2: value += 20 return value def MCTS(board, player, depth): global curAttempts legal = GetLegalPositions(board) value = [] for i in legal: v = ValueFunc(board,player,i) value.append(v) value = np.array(value, dtype=np.float) value = np.exp(value) / np.sum(np.exp(value)) if (value.shape[0] == 0): return 0, 1 id = Serialize(board) if (id not in record): record[id] = [0, 0, 0] localTot = 0 localWin = 0 for i in range(math.ceil(attempts / (depth+1))): curAttempts += 1 if curAttempts > totAttempts: break win, x, y = Step(board, player, (lambda board, player: RandomStep(legal, value)) ) if (win == 0): childWin, childTot = MCTS(board, 3-player, depth+1) record[id][player] += childTot - childWin record[id][3-player] += childWin record[id][0] += childTot localWin += childTot - childWin localTot += childTot else: record[id][win] += 1 record[id][0] += 1 localWin += 1 if win == player else 0 localTot += 1 board[x, y] = EMPTY return (localWin, localTot) def AIPlayer(board, player): global curAttempts curAttempts = 0 print("AI thinking...") if (CurStep(board) == 0): return (board.shape[0])//2, (board.shape[1])//2 MCTS(board, player, 0) legal = GetLegalPositions(board) maxi = -1 maxpos = (0, 0) for i in legal: board[i[0], i[1]] = player id = Serialize(board) if id not in record: board[i[0], i[1]] = EMPTY continue item = record[id] if (item[player] / item[0] >= maxi): maxi = item[player] / item[0] maxpos = i board[i[0], i[1]] = EMPTY print("Max pos =", maxpos, ", prob. of win:", maxi) return maxpos