-
Notifications
You must be signed in to change notification settings - Fork 0
/
8_train_attack_models.py
219 lines (193 loc) · 10.6 KB
/
8_train_attack_models.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
from sklearn import model_selection
from sklearn.naive_bayes import GaussianNB
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.neural_network import MLPClassifier
from sklearn.ensemble import GradientBoostingClassifier
import numpy as np
import pickle
import sys
from imblearn.over_sampling import RandomOverSampler, SMOTE, ADASYN, BorderlineSMOTE
from utils import load_data_csv, fix_randomness
import yaml
fix_randomness()
# read config file
with open("./config/config.yaml", "r") as ymlfile:
try:
cfg = yaml.safe_load(ymlfile)
except yaml.YAMLError as exc:
print(exc)
problem = cfg['problem']
# train on X_train data
X = load_data_csv(cfg["dataset"][problem]["path_to_x_train"])
y = load_data_csv(cfg["dataset"][problem]["path_to_y_train"])
# balance the dataset
if cfg['attack_model']['balance_data']['method'] == 'None':
pass
elif cfg['attack_model']['balance_data']['method'] == 'RandomOverSampler':
X, y = RandomOverSampler(random_state=0).fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'SMOTE':
X, y = SMOTE().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'ADASYN':
X, y = ADASYN().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'BorderlineSMOTE':
X, y = BorderlineSMOTE().fit_resample(X, y)
else:
print("Balance data method is not supported!!!")
sys.exit()
X_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, y, test_size=cfg["attack_model"]["test_size"], random_state=cfg["random"]["random_state_sklearn"])
print("Training {} attack model".format(cfg["attack_model"]["model"]))
if cfg["attack_model"]["model"] == "KNN":
model = KNeighborsClassifier(n_neighbors=cfg["attack_model"]["KNN"]["n_neighbors"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["KNN"]["model_path"], 'wb'))
elif cfg["attack_model"]["model"] == "NB":
model = GaussianNB()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["NB"]["model_path"], 'wb'))
elif cfg["attack_model"]["model"] == "LR":
model = LogisticRegression(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["LR"]["model_path"], 'wb'))
elif cfg["attack_model"]["model"] == "DT":
model = DecisionTreeClassifier(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["DT"]["model_path"], 'wb'))
elif cfg["attack_model"]["model"] == "MLP":
model = MLPClassifier(solver=cfg["attack_model"]["MLP"]["solver"], alpha=1e-5, hidden_layer_sizes=(6,2), random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["MLP"]["model_path"], 'wb'))
elif cfg["attack_model"]["model"] == "XGBoost":
model = GradientBoostingClassifier()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["XGBoost"]["model_path"], 'wb'))
# train on X_train data (with noise)
X = load_data_csv(cfg["dataset"][problem]["path_to_x_train_with_noise"])
y = load_data_csv(cfg["dataset"][problem]["path_to_y_train"])
# balance the dataset
if cfg['attack_model']['balance_data']['method'] == 'None':
pass
elif cfg['attack_model']['balance_data']['method'] == 'RandomOverSampler':
X, y = RandomOverSampler(random_state=0).fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'SMOTE':
X, y = SMOTE().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'ADASYN':
X, y = ADASYN().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'BorderlineSMOTE':
X, y = BorderlineSMOTE().fit_resample(X, y)
else:
print("Balance data method is not supported!!!")
sys.exit()
X_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, y, test_size=cfg["attack_model"]["test_size"], random_state=cfg["random"]["random_state_sklearn"])
print("Training {} attack model (with noise data)".format(cfg["attack_model"]["model"]))
if cfg["attack_model"]["model"] == "KNN":
model = KNeighborsClassifier(n_neighbors=cfg["attack_model"]["KNN"]["n_neighbors"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["KNN"]["model_path_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "NB":
model = GaussianNB()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["NB"]["model_path_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "LR":
model = LogisticRegression(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["LR"]["model_path_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "DT":
model = DecisionTreeClassifier(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["DT"]["model_path_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "MLP":
model = MLPClassifier(solver=cfg["attack_model"]["MLP"]["solver"], alpha=1e-5, hidden_layer_sizes=(6,2), random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["MLP"]["model_path_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "XGBoost":
model = GradientBoostingClassifier()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["XGBoost"]["model_path_with_noise"], 'wb'))
# train on X_train data (with DP)
X = load_data_csv(cfg["dataset"][problem]["path_to_x_train_with_DP"])
y = load_data_csv(cfg["dataset"][problem]["path_to_y_train"])
# balance the dataset
if cfg['attack_model']['balance_data']['method'] == 'None':
pass
elif cfg['attack_model']['balance_data']['method'] == 'RandomOverSampler':
X, y = RandomOverSampler(random_state=0).fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'SMOTE':
X, y = SMOTE().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'ADASYN':
X, y = ADASYN().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'BorderlineSMOTE':
X, y = BorderlineSMOTE().fit_resample(X, y)
else:
print("Balance data method is not supported!!!")
sys.exit()
X_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, y, test_size=cfg["attack_model"]["test_size"], random_state=cfg["random"]["random_state_sklearn"])
print("Training {} attack model (trained with DP)".format(cfg["attack_model"]["model"]))
if cfg["attack_model"]["model"] == "KNN":
model = KNeighborsClassifier(n_neighbors=cfg["attack_model"]["KNN"]["n_neighbors"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["KNN"]["model_path_with_DP"], 'wb'))
elif cfg["attack_model"]["model"] == "NB":
model = GaussianNB()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["NB"]["model_path_with_DP"], 'wb'))
elif cfg["attack_model"]["model"] == "LR":
model = LogisticRegression(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["LR"]["model_path_with_DP"], 'wb'))
elif cfg["attack_model"]["model"] == "DT":
model = DecisionTreeClassifier(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["DT"]["model_path_with_DP"], 'wb'))
elif cfg["attack_model"]["model"] == "MLP":
model = MLPClassifier(solver=cfg["attack_model"]["MLP"]["solver"], alpha=1e-5, hidden_layer_sizes=(6,2), random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["MLP"]["model_path_with_DP"], 'wb'))
elif cfg["attack_model"]["model"] == "XGBoost":
model = GradientBoostingClassifier()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["XGBoost"]["model_path_with_DP"], 'wb'))
# train on X_train data (with DP with noise)
X = load_data_csv(cfg["dataset"][problem]["path_to_x_train_with_DP_with_noise"])
y = load_data_csv(cfg["dataset"][problem]["path_to_y_train"])
# balance the dataset
if cfg['attack_model']['balance_data']['method'] == 'None':
pass
elif cfg['attack_model']['balance_data']['method'] == 'RandomOverSampler':
X, y = RandomOverSampler(random_state=0).fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'SMOTE':
X, y = SMOTE().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'ADASYN':
X, y = ADASYN().fit_resample(X, y)
elif cfg['attack_model']['balance_data']['method'] == 'BorderlineSMOTE':
X, y = BorderlineSMOTE().fit_resample(X, y)
else:
print("Balance data method is not supported!!!")
sys.exit()
X_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, y, test_size=cfg["attack_model"]["test_size"], random_state=cfg["random"]["random_state_sklearn"])
print("Training {} attack model (trained with DP) with noise data".format(cfg["attack_model"]["model"]))
if cfg["attack_model"]["model"] == "KNN":
model = KNeighborsClassifier(n_neighbors=cfg["attack_model"]["KNN"]["n_neighbors"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["KNN"]["model_path_with_DP_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "NB":
model = GaussianNB()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["NB"]["model_path_with_DP_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "LR":
model = LogisticRegression(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["LR"]["model_path_with_DP_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "DT":
model = DecisionTreeClassifier(random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["DT"]["model_path_with_DP_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "MLP":
model = MLPClassifier(solver=cfg["attack_model"]["MLP"]["solver"], alpha=1e-5, hidden_layer_sizes=(6,2), random_state=cfg["random"]["random_state_sklearn"])
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["MLP"]["model_path_with_DP_with_noise"], 'wb'))
elif cfg["attack_model"]["model"] == "XGBoost":
model = GradientBoostingClassifier()
model.fit(X_train, Y_train)
pickle.dump(model, open(cfg["attack_model"]["XGBoost"]["model_path_with_DP_with_noise"], 'wb'))