-
Notifications
You must be signed in to change notification settings - Fork 0
/
newtuner.py
235 lines (201 loc) · 7.72 KB
/
newtuner.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
227
228
229
230
231
232
233
234
235
from __future__ import division
import random, pdb
from run import writefile
from sklearn.cross_validation import StratifiedKFold
# from base import *
from newabcd import sk_abcd
class deBase(object):
def __init__(self, predictor, tuned_objective, train_X, train_Y, test_X,
test_Y, file_name):
global The
self.tobetuned = predictor.tunelst
self.limit_max = predictor.tune_max
self.limit_min = predictor.tune_min
self.predictor = predictor
self.train_X = train_X
self.train_Y = train_Y
self.test_X = test_X
self.test_Y = test_Y
self.np = 10
self.fa = 0.75
self.cr = 0.3
self.repeats = 50
self.life = 5
self.obj = tuned_objective
self.file_name = file_name
self.evaluation = 0
self.scores = {}
self.frontier = [self.generate() for _ in xrange(self.np)]
self.evaluate()
self.bestconf, self.bestscore = self.best()
def generate(self):
candidates = []
for n, item in enumerate(self.limit_min):
if isinstance(item, float):
candidates.append(
round(random.uniform(self.limit_min[n], self.limit_max[n]),
2))
elif isinstance(item, bool):
candidates.append(random.random() <= 0.5)
elif isinstance(item, list):
pass
elif isinstance(item, int):
candidates.append(
int(random.uniform(self.limit_min[n], self.limit_max[n])))
else:
raise ValueError("type of limits are wrong!")
# pdb.set_trace()
return self.treat(candidates)
def evaluate(self):
for n, arglst in enumerate(self.frontier):
# clf = self.assign(arglst)
clf, threshold = self.assign(arglst)
self.scores[n] = self.callModel(clf, threshold)
# main return [[pd,pf,prec,f,g],[pd,pf,prec,f,g]], which are
# N-defective,Y-defecitve
def assign(self, tunedvalue):
param_dict = {}
threshold = None
for key, val in zip(self.tobetuned, tunedvalue):
if key != "threshold":
param_dict[key] = val
else:
threshold = val
param_dict["random_state"] = 1
clf = self.predictor.default(param_dict).fit(self.train_X,
self.train_Y)
return clf, threshold
def best(self):
sortlst = []
if self.obj == 1: # this is for pf
sortlst = sorted(self.scores.items(), key=lambda x: x[1][self.obj],
reverse=True) # alist of turple
else:
sortlst = sorted(self.scores.items(),
key=lambda x: x[1][self.obj]) # alist of turple
bestconf = self.frontier[
sortlst[-1][0]] # [(0, [100, 73, 9, 42]), (1, [75, 41, 12, 66])]
bestscore = sortlst[-1][-1][self.obj]
return bestconf, bestscore
def callModel(self, clf, threshold):
predict_result = clf.predict(self.test_X)
# predict_pro = clf.predict_proba(self.test_X)
scores = sk_abcd(predict_result, self.test_Y, threshold)
return scores[-1]
def treat(self, lst):
"""
some parameters may have constraints, for example:
when generating a parameter list, p[4]should be greater than p[5]
You should implement this function in subclass
"""
# return NotImplementedError("treat error")
return lst
def trim(self, n, x):
if isinstance(self.limit_min[n], float):
return max(self.limit_min[n], min(round(x, 2), self.limit_max[n]))
elif isinstance(self.limit_max[n], int):
return max(self.limit_min[n], min(int(x), self.limit_max[n]))
else:
raise ValueError("wrong type here in parameters")
def gen3(self, n, f):
seen = [n]
def gen1(seen):
while 1:
k = random.randint(0, self.np - 1)
if k not in seen:
seen += [k]
break
return self.frontier[k]
a = gen1(seen)
b = gen1(seen)
c = gen1(seen)
return a, b, c
def update(self, index, old):
newf = []
a, b, c = self.gen3(index, old)
for k in xrange(len(old)):
if isinstance(self.limit_min[k], bool):
newf.append(
old[k] if self.cr < random.random() else not old[k])
elif isinstance(self.limit_min[k], list):
pass
else:
newf.append(
old[k] if self.cr < random.random() else self.trim(k, (
a[k] + self.fa * (b[k] - c[k]))))
return self.treat(newf)
def writeResults(self):
# for p in self.tobetuned:
# temp = 0
# # exec ("temp =" + p)
# writefile(self.file_name, p + ": " + str(temp))
writefile(self.file_name, "evaluation: " + str(self.evaluation))
writefile(self.file_name, "final bestescore: " + str(self.bestscore))
writefile(self.file_name, "final config:" + str(self.bestconf))
def DE(self):
changed = False
def isBetter(new, old):
return new < old if self.obj == 1 else new > old
for k in xrange(self.repeats):
if self.life <= 0:
break
nextgeneration = []
for index, f in enumerate(self.frontier):
new = self.update(index, f)
clf, threshold = self.assign(new)
newscore = self.callModel(clf, threshold)
self.evaluation += 1
if isBetter(newscore[self.obj], self.scores[index][self.obj]):
nextgeneration.append(new)
self.scores[index] = newscore[:]
else:
nextgeneration.append(f)
self.frontier = nextgeneration[:]
newbestconf, newbestscore = self.best()
if isBetter(newbestscore, self.bestscore):
# print "newbestscore %s:" % str(newbestscore)
# print "bestconf %s :" % str(newbestconf)
self.bestscore = newbestscore
self.bestconf = newbestconf[:]
changed = True
if not changed:
self.life -= 1
changed = False
self.writeResults()
print "final bestescore %s: " + str(self.bestscore)
print "final bestconf %s: " + str(self.bestconf)
print "DONE !!!!"
clf,threshold = self.assign(self.bestconf)
return clf,threshold
class WhereDE(deBase):
def __init__(self, predictor):
super(WhereDE, i).__init__(predictor)
def treat(self, lst):
"""
The.where.depthmin < depthMax
"""
def ig(l): return int(
random.uniform(self.limit_min[l], self.limit_max[l]))
if lst[-1] and lst[4] <= lst[5]:
lst[4] = ig(4)
lst[5] = ig(5)
lst = self.treat(lst)
return lst
class CartDE(deBase):
def __init__(self, predictor):
super(CartDE, self).__init__(predictor)
def treat(self, lst):
return lst
class RfDE(deBase):
def __init__(self, predictor):
super(RfDE, self).__init__(predictor)
def treat(self, lst):
return lst
def DE_tuner(predictor, goal_index, new_train_X, new_train_Y, new_test_X,
new_test_Y, file_name):
tuner = deBase(predictor, goal_index, new_train_X, new_train_Y, new_test_X,
new_test_Y, file_name)
clf = tuner.DE()
return clf
if __name__ == "__main__":
Where().DE()