-
Notifications
You must be signed in to change notification settings - Fork 0
/
automl_m2.py
283 lines (215 loc) · 10.7 KB
/
automl_m2.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
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
280
281
282
283
# -*- coding: utf-8 -*-
"""
Created on Tue Oct 29 12:11:11 2019
@author: fvice
"""
# -*- coding: utf-8 -*-
"""
Created on Fri Oct 25 15:42:29 2019
@author: fvice
"""
"""model"""
import numpy as np
import pandas as pd
from hyperopt import STATUS_OK, Trials, hp, space_eval, tpe, fmin
import lightgbm as lgb
from sklearn.model_selection import train_test_split
from util import log
from preprocess import sample
class AutoSSLClassifier:
def __init__(self, train_time_budget):
self.iter = 5
self.label_data = 500
self.model = None
self.train_time_budget = train_time_budget
self.multiplier = 1
def fit(self, X, y):
params = {
"objective": "binary",
"metric": "auc",
"verbosity": -1,
"seed": 1,
"num_threads": 4
}
X_label, y_label, X_unlabeled, y_unlabeled = self._split_by_label(X, y)
y_n_cnt, y_p_cnt = y_label.value_counts()
dims = X_label.shape[0] * X_label.shape[1]
#self.multiplier = min(max(int(self.train_time_budget * 100/(dims**0.75 * np.log(dims))), 1), 50)
self.multiplier = min(max(int(1000 * 100000/(dims * np.log(dims))), 1), 50)
print(self.multiplier)
self.label_data = min(y_n_cnt, y_p_cnt)
y_n = max(int(self.label_data*(y_n_cnt*1.0/len(y_label))), 1)
y_p = max(int(self.label_data*(y_p_cnt*1.0/len(y_label))), 1)
for _ in range(self.iter * max(int(self.multiplier/10), 1)):
if X_unlabeled.shape[0] < self.label_data:
break
hyperparams = self._hyperopt(X_label, y_label, params)
X_train, X_val, y_train, y_val = train_test_split(X_label, y_label, test_size=0.15, random_state=1)
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
self.model = lgb.train({**params, **hyperparams}, train_data, 400 * max(int(self.multiplier/4), 1),
valid_data, early_stopping_rounds= 20 * max(int(self.multiplier/4), 1), verbose_eval=0)
y_hat = self.model.predict(X_unlabeled)
if len(set(y_hat)) == 1:
break
idx = np.argsort(y_hat)
y_p_idx = idx[-y_p:]
y_n_idx = idx[:y_n]
X_label = pd.concat([X_label, X_unlabeled.iloc[list(y_p_idx) + list(y_n_idx), :]])
y_label = pd.concat([y_label, pd.Series([1]*len(y_p_idx) + [-1]*len(y_n_idx))])
X_unlabeled = X_unlabeled.iloc[idx[y_n:-y_p], :]
return self
def predict(self, X):
return self.model.predict(X)
def _split_by_label(self, X, y):
y_label = pd.concat([y[y == -1], y[y == 1]])
X_label = X.loc[y_label.index, :]
y_unlabeled = y[y == 0]
X_unlabeled = X.loc[y_unlabeled.index, :]
return X_label, y_label, X_unlabeled, y_unlabeled
def _hyperopt(self, X, y, params):
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.5, random_state=1)
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
space = {
"learning_rate": hp.loguniform("learning_rate", np.log(0.01), np.log(0.5)),
"max_depth": hp.choice("max_depth", [2, 3, 4, 5, 6]),
"num_leaves": hp.choice("num_leaves", np.linspace(50, 200, 11, dtype=int)),
"feature_fraction": hp.quniform("feature_fraction", 0.6, 1.0, 0.1),
"reg_alpha": hp.uniform("reg_alpha", 0, 2),
"reg_lambda": hp.uniform("reg_lambda", 0, 2),
"min_child_weight": hp.uniform('min_child_weight', 0.5, 10),
}
def objective(hyperparams):
model = lgb.train({**params, **hyperparams}, train_data, 100 * self.multiplier,
valid_data, early_stopping_rounds= 20 * max(int(self.multiplier/2), 1), verbose_eval=0)
score = model.best_score["valid_0"][params["metric"]]
return {'loss': -score, 'status': STATUS_OK}
trials = Trials()
best = fmin(fn=objective, space=space, trials=trials,
algo=tpe.suggest, max_evals=2 * max(int(self.multiplier/2), 1), verbose=0,
rstate=np.random.RandomState(1))
hyperparams = space_eval(space, best)
log(f"auc = {-trials.best_trial['result']['loss']:0.4f} {hyperparams}")
return hyperparams
class AutoPUClassifier:
def __init__(self, train_time_budget):
self.iter = 2
self.models = []
self.train_time_budget = train_time_budget
self.multiplier = 1
def fit(self, X, y):
params = {
"objective": "binary",
"metric": "auc",
"verbosity": -1,
"seed": 1,
"num_threads": 4
}
sample_size = 40000
dims = sample_size * X.shape[1]
self.multiplier = min(max(int(1000 * 100000/(dims * np.log(dims))), 1), 50)
print(self.multiplier)
for _ in range(self.iter * self.multiplier):
x_sample, y_sample = self._negative_sample(X, y)
X_sample, y_sample = sample(x_sample, y_sample, sample_size)
hyperparams = self._hyperopt(X_sample, y_sample, params)
X_train, X_val, y_train, y_val = train_test_split(X_sample, y_sample, test_size=0.15, random_state=1)
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
model = lgb.train({**params, **hyperparams}, train_data, 400 * max(int(self.multiplier/4), 1),
valid_data, early_stopping_rounds=20 * max(int(self.multiplier/4), 1), verbose_eval=0)
self.models.append(model)
return self
def predict(self, X):
for idx, model in enumerate(self.models):
p = model.predict(X)
if idx == 0:
prediction = p
else:
prediction = np.vstack((prediction, p))
return np.mean(prediction, axis=0)
def _negative_sample(self, X, y):
y_n_cnt, y_p_cnt = y.value_counts()
y_n_sample = y_p_cnt if y_n_cnt > y_p_cnt else y_n_cnt
y_sample = pd.concat([y[y == 0].sample(y_n_sample), y[y == 1]])
x_sample = X.loc[y_sample.index, :]
return x_sample, y_sample
def _hyperopt(self, X, y, params):
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.5, random_state=1)
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
space = {
"learning_rate": hp.loguniform("learning_rate", np.log(0.01), np.log(0.5)),
"max_depth": hp.choice("max_depth", [3, 4, 5, 6]),
"num_leaves": hp.choice("num_leaves", np.linspace(50, 200, 11, dtype=int)),
"feature_fraction": hp.quniform("feature_fraction", 0.6, 1.0, 0.2),
"bagging_fraction": hp.quniform("bagging_fraction", 0.4, 1.0, 0.2),
"bagging_freq": hp.choice("bagging_freq", np.linspace(0, 50, 10, dtype=int)),
"reg_alpha": hp.uniform("reg_alpha", 0, 2),
"reg_lambda": hp.uniform("reg_lambda", 0, 2),
"min_child_weight": hp.uniform('min_child_weight', 0.5, 10),
}
def objective(hyperparams):
model = lgb.train({**params, **hyperparams}, train_data, 100 * self.multiplier,
valid_data, early_stopping_rounds=20 * max(int(self.multiplier/2), 1), verbose_eval=0)
score = model.best_score["valid_0"][params["metric"]]
return {'loss': -score, 'status': STATUS_OK}
trials = Trials()
best = fmin(fn=objective, space=space, trials=trials,
algo=tpe.suggest, max_evals=2 * max(int(self.multiplier/4), 1), verbose=0,
rstate=np.random.RandomState(1))
hyperparams = space_eval(space, best)
log(f"auc = {-trials.best_trial['result']['loss']:0.4f} {hyperparams}")
return hyperparams
class AutoNoisyClassifier:
def __init__(self, train_time_budget):
self.model = None
self.train_time_budget = train_time_budget
self.multiplier = 1
def fit(self, X, y):
params = {
"objective": "binary",
"metric": "auc",
"verbosity": -1,
"seed": 1,
"num_threads": 4
}
dims = X.shape[0] * X.shape[1]
self.multiplier = min(max(int(1000 * 100000/(dims * np.log(dims))), 1), 50)
print(self.multiplier)
X_sample, y_sample = sample(X, y, len(X))
hyperparams = self._hyperopt(X_sample, y_sample, params)
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.15, random_state=1)
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
self.model = lgb.train({**params, **hyperparams}, train_data, 500 * max(int(self.multiplier/4), 1),
valid_data, early_stopping_rounds= 20 * max(int(self.multiplier/4), 1), verbose_eval=0)
return self
def predict(self, X):
return self.model.predict(X)
def _hyperopt(self, X, y, params):
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.5, random_state=1)
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
space = {
"learning_rate": hp.loguniform("learning_rate", np.log(0.01), np.log(0.5)),
"max_depth": hp.choice("max_depth", [3, 4, 5, 6, 7]),
"num_leaves": hp.choice("num_leaves", np.linspace(50, 200, 11, dtype=int)),
"feature_fraction": hp.quniform("feature_fraction", 0.5, 1.0, 0.1),
"reg_alpha": hp.uniform("reg_alpha", 0, 2),
"reg_lambda": hp.uniform("reg_lambda", 0, 2),
"min_child_weight": hp.uniform('min_child_weight', 0.5, 10),
}
def objective(hyperparams):
model = lgb.train({**params, **hyperparams}, train_data, 300 * max(int(self.multiplier/2), 1),
valid_data, early_stopping_rounds=20 * max(int(self.multiplier/2), 1), verbose_eval=0)
score = model.best_score["valid_0"][params["metric"]]
return {'loss': -score, 'status': STATUS_OK}
trials = Trials()
best = fmin(fn=objective, space=space, trials=trials,
algo=tpe.suggest, max_evals= 2 * max(int(self.multiplier/2), 1), verbose=0,
rstate=np.random.RandomState(1))
hyperparams = space_eval(space, best)
log(f"auc = {-trials.best_trial['result']['loss']:0.4f} {hyperparams}")
return hyperparams