-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathattacking_to_models.py
204 lines (151 loc) · 9.33 KB
/
attacking_to_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
from misc.data_loader import DataLoader
from misc.models import MODELS
from misc.attacks import fgsm_attack, pgd_attack
from misc.evaluation_metrics import *
from tensorflow.keras.models import load_model
from tqdm import tqdm
import json
import warnings
warnings.filterwarnings("ignore")
import argparse
parser = argparse.ArgumentParser(
prog = "Vanilla Model Training",
description = "Trains the models defined in 'models.py' file and saves them to 'models/'"
)
parser.add_argument("--setting", required=True)
parser.add_argument("--attack", required=True)
parser.add_argument("--models")
args = parser.parse_args()
settings = args.setting.lower()
attacks = args.attack.lower()
settings = [item.strip() for item in settings.split(",")]
attacks = [item.strip() for item in attacks.split(",")]
dl = DataLoader()
info = dl.get_info()
for setting in settings:
if setting == "mto":
with open("models/many-to-one/model_info.json", "r") as f:
model_info = json.loads(f.read())
info = info[info.name.isin(list(model_info.keys()))]
adv_samples = {}
for index, row in tqdm(info.iterrows(), desc="dataset", position=0) :
adv_samples[row["name"]] = {}
df = dl.load(index)
if row["name"] == "Electricity Transformer Data - 15 min":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(23*4,1)
elif row["name"] == "Metro Interstate Human Traffic Volume":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(23,1)
elif row["name"] == "Beijing-Guanyuan Air-Quality":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(23,1)
elif row["name"] == "Solar Generation - EnerjiSA":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(23,1)
for model_name in tqdm(model_info[row["name"]].keys(), desc="models", position=1):
adv_samples[row["name"]][model_name] = {"FGSM": {}, "PGD": {}}
model_path = model_info[row["name"]][model_name]["path"]
model = load_model(model_path)
if "fgsm" in attacks:
# FGSM attack
for epsilon in tqdm([0.01, 0.025, 0.05, 0.1], desc="FGSM", position=2):
X_adv = fgsm_attack(X_test, y_test, model, epsilon, np.inf, clip_min=0, clip_max=100000)
pred = model.predict(X_adv)
pred = scaler.inverse_transform(pred).reshape(1,-1)[0]
y_test_inv = scaler.inverse_transform(y_test).reshape(1,-1)[0]
adv_samples[row["name"]][model_name]["FGSM"]["epsilon=" + str(epsilon)] = {
# "data" : X_adv.numpy().reshape(X_adv.shape[0],X_adv.shape[1]),
"metrics" : {
"R2" : round(r2_score(y_test_inv,pred),3),
"MAE" : round(MAE(y_test_inv,pred),2),
"RMSE" : round(RMSE(y_test_inv,pred),2),
"MSE" : round(MSE(y_test_inv,pred),2),
"MAPE" : round(MAPE(y_test_inv,pred),2),
"SMAPE" : round(SMAPE(y_test_inv,pred),2),
"MDAPE" : round(MDAPE(y_test_inv,pred),2)
}
}
if "pgd" in attacks:
# PGD attack
for alpha in tqdm([0.01, 0.025], desc="PGD", position=2):
for epsilon in [0.01, 0.025]:
iterations = 7
X_adv = pgd_attack(X_test, y_test, model, iterations, alpha, epsilon, np.inf, clip_min=0, clip_max=100000)
pred = model.predict(X_adv)
pred = scaler.inverse_transform(pred).reshape(1,-1)[0]
y_test_inv = scaler.inverse_transform(y_test).reshape(1,-1)[0]
adv_samples[row["name"]][model_name]["PGD"]["alpha=" + str(alpha) + " | epsilon=" + str(epsilon)] = {
# "data" : str(X_adv),
"metrics" : {
"R2" : round(r2_score(y_test_inv,pred),3),
"MAE" : round(MAE(y_test_inv,pred),2),
"RMSE" : round(RMSE(y_test_inv,pred),2),
"MSE" : round(MSE(y_test_inv,pred),2),
"MAPE" : round(MAPE(y_test_inv,pred),2),
"SMAPE" : round(SMAPE(y_test_inv,pred),2),
"MDAPE" : round(MDAPE(y_test_inv,pred),2)
}
}
with open("adv_examples/many-to-one/adv_gen_l_inf.json", "w") as outfile:
json.dump(adv_samples, outfile)
print("Model performances against adversarial attacks have been saved to 'adv_examples/many-to-one/adv_gen_l_inf.json'")
elif setting == "mtm":
with open("models/many-to-many/model_info.json", "r") as f:
model_info = json.loads(f.read())
info = info[info.name.isin(list(model_info.keys()))]
adv_samples = {}
for index, row in tqdm(info.iterrows(), desc="dataset", position=0) :
adv_samples[row["name"]] = {}
df = dl.load(index)
if row["name"] == "Electricity Transformer Data - 15 min":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(24*7,12)
elif row["name"] == "Metro Interstate Human Traffic Volume":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(24*7,12)
elif row["name"] == "Beijing-Guanyuan Air-Quality":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(24*7,12)
elif row["name"] == "Solar Generation - EnerjiSA":
(X_train, y_train, X_test, y_test), scaler = dl.prepare_dataset(24*7,12)
for model_name in tqdm(model_info[row["name"]].keys(), desc="models", position=1):
adv_samples[row["name"]][model_name] = {"FGSM": {}, "PGD": {}}
model_path = model_info[row["name"]][model_name]["path"]
model = load_model(model_path)
if "fgsm" in attacks:
# FGSM attack
for epsilon in tqdm([0.01, 0.025, 0.05, 0.1], desc="FGSM", position=2):
X_adv = fgsm_attack(X_test, y_test, model, epsilon, np.inf, clip_min=0, clip_max=100000)
pred = model.predict(X_adv)
pred = scaler.inverse_transform(pred).reshape(1,-1)[0]
y_test_inv = scaler.inverse_transform(y_test).reshape(1,-1)[0]
adv_samples[row["name"]][model_name]["FGSM"]["epsilon=" + str(epsilon)] = {
# "data" : X_adv.numpy().reshape(X_adv.shape[0],X_adv.shape[1]),
"metrics" : {
"R2" : round(r2_score(y_test_inv,pred),3),
"MAE" : round(MAE(y_test_inv,pred),2),
"RMSE" : round(RMSE(y_test_inv,pred),2),
"MSE" : round(MSE(y_test_inv,pred),2),
"MAPE" : round(MAPE(y_test_inv,pred),2),
"SMAPE" : round(SMAPE(y_test_inv,pred),2),
"MDAPE" : round(MDAPE(y_test_inv,pred),2)
}
}
if "pgd" in attacks:
# PGD attack
for alpha in tqdm([0.01, 0.025], desc="PGD", position=2):
for epsilon in [0.01, 0.025]:
iterations = 7
X_adv = pgd_attack(X_test, y_test, model, iterations, alpha, epsilon, np.inf, clip_min=0, clip_max=100000)
pred = model.predict(X_adv)
pred = scaler.inverse_transform(pred).reshape(1,-1)[0]
y_test_inv = scaler.inverse_transform(y_test).reshape(1,-1)[0]
adv_samples[row["name"]][model_name]["PGD"]["alpha=" + str(alpha) + " | epsilon=" + str(epsilon)] = {
# "data" : str(X_adv),
"metrics" : {
"R2" : round(r2_score(y_test_inv,pred),3),
"MAE" : round(MAE(y_test_inv,pred),2),
"RMSE" : round(RMSE(y_test_inv,pred),2),
"MSE" : round(MSE(y_test_inv,pred),2),
"MAPE" : round(MAPE(y_test_inv,pred),2),
"SMAPE" : round(SMAPE(y_test_inv,pred),2),
"MDAPE" : round(MDAPE(y_test_inv,pred),2)
}
}
with open("adv_examples/many-to-many/adv_gen_l_inf.json", "w") as outfile:
json.dump(adv_samples, outfile)
print("Model performances against adversarial attacks have been saved to 'adv_examples/many-to-many/adv_gen_l_inf.json'")