-
Notifications
You must be signed in to change notification settings - Fork 1
/
monte_carlo_tree.py
180 lines (154 loc) · 5.94 KB
/
monte_carlo_tree.py
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
import math
import copy
import random
import numpy as np
from baseClass.Player_class import Player
class Node(object):
def __init__(self, parent, borad_2d, action):
self.parent = parent
self.borad = copy.deepcopy(borad_2d)
self.childrens = {}
self.win = 0
self.visits = 1
self.reward = 0
self.action = action
def __str__(self):
return str(self.borad)
def isFullExpand(self):
for v in self.childrens.values():
if v.reward == 0:
return False
return True
def isExpend(self):
if len(self.childrens) != 0:
return True
else:
return False
def addChildren(self, node):
node_id = tuple( map(tuple, node.borad) )
self.childrens[node_id] = node
# --------------------------------------------------------------------------
class Monte_Carlo_Player(Player):
def __init__(self, borad):
self.gameBorad = borad
self.root = Node(None, self.gameBorad.borad, None)
self.currentNode = None
def init_game(self):
self.gameBorad.startGame()
# self.gameBorad.start_FakeGame()
self.currentNode = None
def action(self):
if self.currentNode == None:
self.currentNode = self.root
key = tuple( map(tuple, self.gameBorad.borad))
if key in self.currentNode.childrens:
self.currentNode = self.currentNode.childrens[key]
else:
newChild = Node(self.currentNode, self.gameBorad.borad, None)
self.currentNode.childrens[key] = newChild
self.currentNode = newChild
bestChild = self.uctSearch(self.currentNode)
self.currentNode.action = bestChild.action # Record current borad with chosen action
self.currentNode = bestChild
self.gameBorad.restore_borad_info()
bestChild.action(self.gameBorad.simuBorad)
bestChild.action(self.gameBorad.borad)
self.gameBorad.save_borad_info()
return self.currentNode.action.__name__
def uctSearch(self, node):
for i in range(5):
expandNode = self.treePolicy(node)
delta = self.defaultPolicy(expandNode)
self.backpropagation(expandNode, node, delta)
child = self.bestChild(node)
return child
def treePolicy(self, node):
v = node
while not self.gameBorad.isEnd( v.borad ):
if not v.isExpend():
self.expand(v)
if v.isFullExpand():
v = self.bestChild(v)
else:
notExpandNode = [ c for c in v.childrens.values() if c.reward == 0 ]
return random.choice(notExpandNode)
return v
def expand(self, v):
for nextAction in self.gameBorad.actionTable:
simuBorad = copy.deepcopy(v.borad)
succ_move = nextAction(simuBorad)
if succ_move:
n = Node(v, simuBorad, nextAction)
v.addChildren(n)
self.gameBorad.restore_borad_info()
def bestChild(self, node):
argmax = -1
argNode = None
for v in node.childrens.values():
score = v.reward/v.visits + 1.414*math.sqrt( 2*math.log(node.visits) / v.visits)
if score > argmax:
argmax = score
argNode = v
return argNode
def defaultPolicy(self, v):
reward = 0.0
simulateTimes = 10
if self.gameBorad.isEnd( v.borad ):
reward = self.getFreeReward(v.borad)
return reward
for times in range(simulateTimes):
depth = 20 # if depth == -1, will go through to game end.
self.gameBorad.simuBorad = copy.deepcopy(v.borad)
while depth != 0 and not self.gameBorad.isEnd( self.gameBorad.simuBorad ):
succ_move = False
self.gameBorad.random_blocks(self.gameBorad.simuBorad)
nextAction = self.gameBorad.actionTable[ random.randint(0,3) ]
if nextAction != None:
succ_move = nextAction(self.gameBorad.simuBorad)
if succ_move == True:
reward += self.getFreeReward(self.gameBorad.simuBorad)
depth -= 1
self.gameBorad.restore_borad_info()
return reward / simulateTimes
'''
Why there are children with same borad???
'''
def backpropagation(self, v, node, delta):
while v != node:
v.visits += 1
v.reward += delta
v = v.parent
else: # parent node should reflash info
v.visits += 1
v.reward += delta
def getFreeReward(self, borad):
score = 0
for i in range(4):
for j in range(4):
'''
if borad[i][j] == 2048:
score = 999999
return score
'''
if i-1 >= 0 and borad[i][j] !=0 and borad[i][j] == borad[i-1][j]:
score += borad[i][j]*2
# score += 1
if i+1 <= 3 and borad[i][j] !=0 and borad[i][j] == borad[i+1][j]:
score += borad[i][j]*2
# score += 1
if j+1 <= 3 and borad[i][j] !=0 and borad[i][j] == borad[i][j+1]:
score += borad[i][j]*2
# score += 1
if j-1 >= 0 and borad[i][j] !=0 and borad[i][j] == borad[i][j-1]:
score += borad[i][j]*2
# score += 1
#------------------------------------------------------------------
if i-1 >= 0 and borad[i-1][j] == 0:
score += 1
if i+1 <= 3 and borad[i+1][j] == 0:
score += 1
if j+1 <= 3 and borad[i][j+1] == 0:
score += 1
if j-1 >= 0 and borad[i][j-1] == 0:
score += 1
return score