-
Notifications
You must be signed in to change notification settings - Fork 0
/
Node.py
226 lines (184 loc) · 7.58 KB
/
Node.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
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
import json
import math
from math import log2, ceil
import random
import numpy as np
from Classifier import Classifier
class Node:
obj_counter = 0
def __init__(self, parent = None, is_good_kid = False, arch_code_len = 0, is_root = False):
# Note: every node is initialized as a leaf,
# only non-leaf nodes are equipped with classifiers to make decisions
if not is_root:
assert type(parent) == type(self)
self.parent = parent
self.is_good_kid = is_good_kid
self.ARCH_CODE_LEN = arch_code_len
self.is_root = is_root
self.x_bar = float("inf")
self.n = 0
self.uct = 0
self.counter = 1
self.kids = []
self.bag = {}
self.validation = {}
self.f1 = [0]
# data for good and bad kids, respectively
self.good_kid_data = {}
self.bad_kid_data = {}
self.is_leaf = True
self.id = Node.obj_counter
self.layer = ceil(log2(self.id + 2) - 1)
self.classifier = Classifier({}, self.ARCH_CODE_LEN, self.id)
# insert current node into the kids of parent
if parent is not None:
self.parent.kids.append(self)
if self.parent.is_leaf == True:
self.parent.is_leaf = False
assert len(self.parent.kids) <= 2
Node.obj_counter += 1
def clear_data(self):
self.bag.clear()
self.bad_kid_data.clear()
self.good_kid_data.clear()
def put_in_bag(self, net, maeinv):
assert type(net) == type([])
# assert type(maeinv) == type(float(0.1))
net_k = json.dumps(net)
self.bag[net_k] = (maeinv)
def get_name(self):
# state is a list of jsons
return "node" + str(self.id)
def pad_str_to_8chars(self, ins):
if len(ins) <= 14:
ins += ' ' * (14 - len(ins))
return ins
else:
return ins
def __str__(self):
name = self.get_name()
name = self.pad_str_to_8chars(name)
name += (self.pad_str_to_8chars('lf:' + str(self.is_leaf)))
name += (self.pad_str_to_8chars(' val:{0:.4f} '.format(round(self.get_xbar(), 4))))
name += (self.pad_str_to_8chars(' uct:{0:.4f} '.format(round(self.get_uct(Cp=0.5), 4))))
name += self.pad_str_to_8chars('n:' + str(self.n))
name += self.pad_str_to_8chars('visit:' + str(self.counter))
if self.is_leaf == False:
name += self.pad_str_to_8chars('acc:{0:.4f} '.format(round(self.classifier.training_accuracy[-1], 4)))
name += self.pad_str_to_8chars('f1:{0:.4f} '.format(round(self.f1[-1], 4)))
else:
name += self.pad_str_to_8chars('acc: ---- ')
name += self.pad_str_to_8chars('f1: ---- ')
name += self.pad_str_to_8chars('sp:' + str(len(self.bag)))
# name += (self.pad_str_to_8chars('g_k:' + str(len(self.good_kid_data))))
# name += (self.pad_str_to_8chars('b_k:' + str(len(self.bad_kid_data))))
parent = '----'
if self.parent is not None:
parent = self.parent.get_name()
parent = self.pad_str_to_8chars(parent)
name += (' parent:' + parent)
# kids = ''
# kid = ''
# for k in self.kids:
# kid = self.pad_str_to_8chars(k.get_name())
# kids += kid
# name += (' kids:' + kids)
if self.is_leaf:
name = Color.YELLOW + name +Color.RESET
elif self.layer == 2:
name = Color.GREEN + name +Color.RESET
return name
def get_uct(self, Cp):
if self.is_root and self.parent == None:
return float('inf')
if self.n == 0:
return float('inf')
# coeff = 2 ** (5 - ceil(log2(self.id + 2)))
if len(self.bag) == 0:
return float('-inf')
# return self.x_bar + Cp*math.sqrt(2*math.log(self.parent.n)/self.n)
return self.x_bar + 2 * Cp*math.sqrt(2*math.log(self.parent.counter)/self.counter)
def get_xbar(self):
return self.x_bar
def train(self):
if self.parent == None and self.is_root == True:
# training starts from the bag
assert len(self.bag) > 0
self.classifier.update_samples(self.bag)
self.good_kid_data, self.bad_kid_data = self.classifier.split_data()
elif self.is_leaf:
if self.is_good_kid:
self.bag = self.parent.good_kid_data
else:
self.bag = self.parent.bad_kid_data
else:
if self.is_good_kid:
self.bag = self.parent.good_kid_data
self.classifier.update_samples(self.parent.good_kid_data)
self.good_kid_data, self.bad_kid_data = self.classifier.split_data()
else:
self.bag = self.parent.bad_kid_data
self.classifier.update_samples(self.parent.bad_kid_data)
self.good_kid_data, self.bad_kid_data = self.classifier.split_data()
if len(self.bag) == 0:
self.x_bar = float('inf')
self.n = 0
else:
self.x_bar = np.mean(np.array(list(self.bag.values())))
self.n = len(self.bag.values())
def predict(self, method = None):
if self.parent == None and self.is_root == True and self.is_leaf == False:
self.good_kid_data, self.bad_kid_data, _ = self.classifier.split_predictions(self.bag, method)
elif self.is_leaf:
if self.is_good_kid:
self.bag = self.parent.good_kid_data
else:
self.bag = self.parent.bad_kid_data
else:
if self.is_good_kid:
self.bag = self.parent.good_kid_data
self.good_kid_data, self.bad_kid_data, xbar = self.classifier.split_predictions(self.parent.good_kid_data, method)
# self.x_bar = xbar
else:
self.bag = self.parent.bad_kid_data
self.good_kid_data, self.bad_kid_data, xbar = self.classifier.split_predictions(self.parent.bad_kid_data, method)
# self.x_bar = xbar
if method:
self.validation = self.bag.copy()
def predict_validation(self):
if self.is_leaf == False:
self.good_kid_data, self.bad_kid_data, _ = self.classifier.split_predictions(self.validation)
if self.is_good_kid:
self.bag = self.parent.good_kid_data
def get_performance(self):
i = 0
for k in self.bag.keys():
if k in self.validation:
i += 1
precision = i / (len(self.bag) + 1e-6)
i = 0
for k in self.validation.keys():
if k in self.bag:
i += 1
recall = i / len(self.validation)
f1 = 2 * precision * recall / (precision + recall + 1e-6)
return f1
def sample_arch(self):
if len(self.bag) == 0:
return None
net_str = random.choice(list(self.bag.keys()))
del self.bag[net_str]
parent_node = self.parent
for i in range(3):
del parent_node.bag[net_str]
parent_node = parent_node.parent
return json.loads(net_str)
class Color:
RED = '\033[31m'
GREEN = '\033[32m'
YELLOW = '\033[33m'
BLUE = '\033[34m'
MAGENTA = '\033[35m'
CYAN = '\033[36m'
WHITE = '\033[37m'
RESET = '\033[0m'