def __init__(self, params):

self.params = params

self.pacmanType = params['pacmanType']

self.ghostType = params['ghostType']

self.initialize(params['layout'])

# self.display = params['display']

# 没回合吃掉豆子的比例

self.rate = 0

# 平均比例

self.avg_rate = 0

self.change = 0

layout = params['layout']

def nextRound(self, layout):

newState = State(layout)

ghost_pos = []

ghosts = []

players = [p for p in self.players if not p.isPacman]

for isPacman, pos, super in layout.agentPositions:

if not isPacman:

players.append(Player(False, (int(pos[0]), int(pos[1])), 0, 0))

self.players = players

self.state.players = players

def createPlayers(self):

players = []

agentParams = {

'num_training': self.params['numTraining'],

'layout': self.params['layout']

}

i = 0

for isPacman, pos, super in self.state.layout.agentPositions:

player = Player(isPacman, (int(pos[0]), int(pos[1])), i, super)

i += 1

player.setAgent(

self.pacmanType if isPacman else self.ghostType, self.state, agentParams)

players.append(player)

# print('createplayers end...')

return players

def initialize(self, layout):

self.state = State(layout)

self.layout = layout

self.players = self.createPlayers()

# print('l....', len(self.players))

self.state.setPlayers(self.players)

# 游戏循环

def run(self):

rnd = 1

for i in range(self.params['numGames']):

self.round(rnd)

rnd += 1

# 每次回合刷新

def refresh(self, rnd):

# rnd = 1000

state = self.state

p_set = set()

RND_NUM = 2000

if rnd % RND_NUM == 0:

# 模拟旋转

# rotate = (random.random())

layoutText = state.layout.layoutText

height = len(layoutText)

width = len(layoutText[0])

newLayout = []

# flip-x

if (rnd // RND_NUM) % 10 in [1,2,6,7]:

print('flip-x')

for i in range(height):

row = ['' for x in range(width)]

for j in range(width):

# print(layoutText[i], j)

row.append(layoutText[i][-1-j])

newLayout.append(row)

newLayoutText = [''.join(p) for p in newLayout]

# flip-y

if (rnd // RND_NUM) % 10 in [3,4,8,9]:

print('flip-y')

for i in range(height):

row = ['' for x in range(width)]

for j in range(width):

row.append(layoutText[-1-i][j])

newLayout.append(row)

newLayoutText = [''.join(p) for p in newLayout]

# rotate

if (rnd // RND_NUM) % 10 in [5, 0]:

print('rotate')

for i in range(width):

row = ['' for x in range(height)]

for j in range(height):

row.append(layoutText[j][i])

newLayout.append(row)

newLayoutText = [''.join(p) for p in newLayout]

# 随机生成pacman和Ghost

# 豆子和空位置的比例

ratio = 0.2 + 0.8*random.random()

for i in range(len(newLayoutText)) :

for j in range(len(newLayoutText[i])):

c = newLayoutText[i][j]

r = random.random()

if c == '.' or c == ' ' or c == 'o' or c == 'P' or c == 'G' :

if r < ratio :

newLayoutText[i] = newLayoutText[i][:j] + '.' + newLayoutText[i][j+1:]

else:

newLayoutText[i] = newLayoutText[i][:j] + ' ' + newLayoutText[i][j+1:]

posList = []

for i in range(len(newLayoutText)) :

for j in range(len(newLayoutText[i])):

c = newLayoutText[i][j]

if c != '%':

posList.append((i, j))

# print(len(posList), len(self.players))

capsuleNum = round( random.random() * 2)

smp = random.sample(posList, len(self.players) + capsuleNum)

agents = []

for i in range(len(smp)) :

y,x = smp[i]

c = ''

if i < len(self.players) :

self.players[i].setInitialPos( (x, len(newLayoutText) - y - 1) )

self.players[i].refresh()

if 'refresh' in dir(self.players[i].agent):

self.players[i].agent.refresh()

c = 'P' if self.players[i].isPacman else 'G'

# agents.append(self.players)

else :

c = 'o'

newLayoutText[y] = newLayoutText[y][:x] + c + newLayoutText[y][x+1:]

newLayout = Layout(newLayoutText)

state.layout = newLayout

state.refresh()

print(state)

# self.initialize(Layout(newLayoutText))

else:

for player in self.players:

player.refresh()

state.refresh()

# 打印游戏结果

def printResult(self, rnd, result, state, eps):

pacmanWin = state.pacmanScore > state.ghostScore

if not pacmanWin:

print('{0}, ghost win, pacman({1}), ghost({2}), eps={3}, rate={4:.2f}'.format(

rnd, self.state.pacmanScore, self.state.ghostScore, eps, self.rate))

else:

print('{0}, pacman win, pacman({1}), ghost({2}), eps={3}, rate={4:.2f}'.format(

rnd, self.state.pacmanScore, self.state.ghostScore, eps, self.rate))

print('avg_rate : %.3f' % self.avg_rate)

# 原来调用Agent的回调方法

def notifyAgentHooks(self, name):

for player in self.players:

if name in dir(player.agent):

method_to_call = getattr(player.agent, name)

method_to_call(self.state)

# player.agent[name](self.state)

# 提示下一步

def hintNextMove(self):

# print('hintNextMove...')print

state = self.state

alivePlayers = state.alivePlayers()

pacmanCount = 0

direct = {

'North': 'UP',

'South': 'DOWN',

'East': 'RIGHT',

'West': 'LEFT',

'Stop': 'STOP'

}

dirs = dict()

for player in alivePlayers:

if not player.alive:

continue

if player.isPacman:

action = player.getAction(state)

state.next(player, action)

dirs[pacmanCount] = direct[action]

pacmanCount += 1

return {'dirs': dirs}

# 每个回合

def round(self, rnd):

result = 0

self.refresh(rnd)

state = self.state

eps = 0

if self.params['numTraining'] < rnd:

self.display = Display()

self.display.initialize(state)

self.notifyAgentHooks('init')

turn = 0

while(True):

players = None

if turn % 2 == 0 :

players = state.alivePacmans()

else :

players = state.aliveGhosts()

for player in players:

action = player.getAction(state)

state.next(player, action)

result = state.whoWins()

if result != 0:

break

if result != 0:

self.notifyAgentHooks('observationFunction')

break

if turn % 2 == 1:

self.notifyAgentHooks('observationFunction')

if self.params['numTraining'] < rnd:

self.display.update(state, rnd)

eps += 1

turn += 1

# self.notifyAgentHooks('observationFunction')

self.notifyAgentHooks('final')

R = max(rnd, 100)

self.rate = 1 - (state.food.count() /

state.total_food) if state.total_food > 0 else 0

self.avg_rate = self.avg_rate * (R-1) / (R) + self.rate / R

# if rnd % 10 == 0:

def initialize(self, state):

print(state)

def update(self, state, rnd):

system('clear')

print('round:', rnd)

print(state)