Spaces:
Runtime error
Runtime error
File size: 8,665 Bytes
6bc2b54 3096d23 6bc2b54 5fbd9b5 6bc2b54 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 |
import numpy as np
import pickle
BOARD_ROWS = 9
BOARD_COLS = 9
class State:
def __init__(self, p1, p2):
self.board = np.zeros((BOARD_ROWS, BOARD_COLS))
self.p1 = p1
self.p2 = p2
self.isEnd = False
self.boardHash = None
# init p1 plays first
self.playerSymbol = 1
# get unique hash of current board state
def getHash(self):
self.boardHash = str(self.board.reshape(BOARD_COLS * BOARD_ROWS))
return self.boardHash
def winner(self):
# row
for i in range(BOARD_ROWS):
if sum(self.board[i, :]) == 3:
self.isEnd = True
return 1
if sum(self.board[i, :]) == -3:
self.isEnd = True
return -1
# col
for i in range(BOARD_COLS):
if sum(self.board[:, i]) == 3:
self.isEnd = True
return 1
if sum(self.board[:, i]) == -3:
self.isEnd = True
return -1
# diagonal
diag_sum1 = sum([self.board[i, i] for i in range(BOARD_COLS)])
diag_sum2 = sum([self.board[i, BOARD_COLS - i - 1] for i in range(BOARD_COLS)])
diag_sum = max(abs(diag_sum1), abs(diag_sum2))
if diag_sum == 3:
self.isEnd = True
if diag_sum1 == 3 or diag_sum2 == 3:
return 1
else:
return -1
# tie
# no available positions
if len(self.availablePositions()) == 0:
self.isEnd = True
return 0
# not end
self.isEnd = False
return None
def availablePositions(self):
positions = []
for i in range(BOARD_ROWS):
for j in range(BOARD_COLS):
if self.board[i, j] == 0:
positions.append((i, j)) # need to be tuple
return positions
def updateState(self, position):
self.board[position] = self.playerSymbol
# switch to another player
self.playerSymbol = -1 if self.playerSymbol == 1 else 1
# only when game ends
def giveReward(self):
result = self.winner()
# backpropagate reward
if result == 1:
self.p1.feedReward(1)
self.p2.feedReward(0)
elif result == -1:
self.p1.feedReward(0)
self.p2.feedReward(1)
else:
self.p1.feedReward(0.1)
self.p2.feedReward(0.5)
# board reset
def reset(self):
self.board = np.zeros((BOARD_ROWS, BOARD_COLS))
self.boardHash = None
self.isEnd = False
self.playerSymbol = 1
def playwithbot(self, rounds=100):
for i in range(rounds):
if i % 1000 == 0:
print("Rounds {}".format(i))
while not self.isEnd:
# Player 1
positions = self.availablePositions()
p1_action = self.p1.chooseAction(positions, self.board, self.playerSymbol)
# take action and upate board state
self.updateState(p1_action)
board_hash = self.getHash()
self.p1.addState(board_hash)
# check board status if it is end
win = self.winner()
if win is not None:
# self.showBoard()
# ended with p1 either win or draw
self.giveReward()
self.p1.reset()
self.p2.reset()
self.reset()
break
else:
# Player 2
positions = self.availablePositions()
p2_action = self.p2.chooseAction(positions, self.board, self.playerSymbol)
self.updateState(p2_action)
board_hash = self.getHash()
self.p2.addState(board_hash)
win = self.winner()
if win is not None:
# self.showBoard()
# ended with p2 either win or draw
self.giveReward()
self.p1.reset()
self.p2.reset()
self.reset()
break
# play with human
def playwithhuman(self):
while not self.isEnd:
# Player 1
positions = self.availablePositions()
p1_action = self.p1.chooseAction(positions, self.board, self.playerSymbol)
# take action and upate board state
self.updateState(p1_action)
self.showBoard()
# check board status if it is end
win = self.winner()
if win is not None:
if win == 1:
print(self.p1.name, "wins!")
else:
print("tie!")
self.reset()
break
else:
# Player 2
positions = self.availablePositions()
p2_action = self.p2.chooseAction(positions)
self.updateState(p2_action)
self.showBoard()
win = self.winner()
if win is not None:
if win == -1:
print(self.p2.name, "wins!")
else:
print("tie!")
self.reset()
break
# def showBoard(self):
# # p1: x p2: o
# for i in range(0, BOARD_ROWS):
# print('-------------')
# out = '| '
# for j in range(0, BOARD_COLS):
# if self.board[i, j] == 1:
# token = 'x'
# if self.board[i, j] == -1:
# token = 'o'
# if self.board[i, j] == 0:
# token = ' '
# out += token + ' | '
# print(out)
# print('-------------')
class Player:
def __init__(self, name, exp_rate=0.3):
self.name = name
self.states = [] # record all positions taken
self.lr = 0.2
self.exp_rate = exp_rate
self.decay_gamma = 0.9
self.states_value = {} # state -> value
self.loadPolicy('policy_' + str(self.name)) # Load the pre-trained policy
def getHash(self, board):
boardHash = str(board.reshape(BOARD_COLS * BOARD_ROWS))
return boardHash
def chooseAction(self, positions, current_board, symbol):
if np.random.uniform(0, 1) <= self.exp_rate:
# take random action
idx = np.random.choice(len(positions))
action = positions[idx]
else:
value_max = -999
for p in positions:
next_board = current_board.copy()
next_board[p] = symbol
next_boardHash = self.getHash(next_board)
value = 0 if self.states_value.get(next_boardHash) is None else self.states_value.get(next_boardHash)
# print("value", value)
if value >= value_max:
value_max = value
action = p
# print("{} takes action {}".format(self.name, action))
return action
# append a hash state
def addState(self, state):
self.states.append(state)
# at the end of game, backpropagate and update states value
def feedReward(self, reward):
for st in reversed(self.states):
if self.states_value.get(st) is None:
self.states_value[st] = 0
self.states_value[st] += self.lr * (self.decay_gamma * reward - self.states_value[st])
reward = self.states_value[st]
def reset(self):
self.states = []
def savePolicy(self):
fw = open('policy_9x9_' + str(self.name), 'wb')
pickle.dump(self.states_value, fw)
fw.close()
def loadPolicy(self, file):
fr = open(file, 'rb')
self.states_value = pickle.load(fr)
fr.close()
class HumanPlayer:
def __init__(self, name):
self.name = name
def chooseAction(self, positions):
pass
# append a hash state
def addState(self, state):
pass
# at the end of game, backpropagate and update states value
def feedReward(self, reward):
pass
def reset(self):
pass
if __name__ == "__main__":
# training
p1 = Player("9x9_p1")
p2 = Player("9x9_p2")
st = State(p1, p2)
print("training...")
st.playwithbot(200000)
p1.savePolicy()
p2.savePolicy()
|