-
Notifications
You must be signed in to change notification settings - Fork 1
/
minibatch_optimize_pt737.py
242 lines (181 loc) · 10.2 KB
/
minibatch_optimize_pt737.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
# -*- coding: utf-8 -*-
"""
Created on Mon Sep 14 15:13:21 2020
@author: 1804499
"""
import numpy as np
from sklearn.model_selection import train_test_split
import memory_profiler
import time
def data():
benign = np.loadtxt("benign_traffic.csv", delimiter = ",")
mirai = np.loadtxt("mirai_traffic.csv", delimiter = ",")
gafgyt = np.loadtxt("gafgyt_traffic.csv", delimiter = ",")
alldata = np.concatenate((benign, gafgyt, mirai))
j = len(benign[0])
data = alldata[:, 1:j]
benlabel = alldata[:, 0]
bendata = (data - data.min()) / (data.max() - data.min())
bendata, benmir, benlabel, benslabel = train_test_split(bendata, benlabel, test_size = 0.2, random_state = 42)
return bendata, benmir, benlabel, benslabel
# implemented densely connected neural network
import fully_connected_nn
#neural network architecture
traind, testd, trainlbl, testlbl = data()
#traind = np.float32(traind)
#testd = np.float32(testd)
#n = len(traind[0])
NN_ARCHITECTURE = [
{"input_dim": 115, "output_dim": 83, "activation": "relu"},
{"input_dim": 83, "output_dim": 83, "activation": "relu"},
{"input_dim": 83, "output_dim":128, "activation": "relu"},
{"input_dim": 128, "output_dim": 83, "activation": "relu"},
{"input_dim": 83, "output_dim": 1, "activation": "sigmoid"}
]
#dataset transformation
traind = np.transpose(traind)
trainlbl = np.transpose(trainlbl.reshape((trainlbl.shape[0], 1)))
testd = np.transpose(testd)
testlbl = np.transpose(testlbl.reshape((testlbl.shape[0], 1)))
N_EPOCHS = 100
LR = 0.001
eps= 0.001
#Baseline model (Basic Gradient)
starttb = time.time()
startmb = memory_profiler.memory_usage()
params_values_b, cost_history_b, accuracy_history_b, grad_a = fully_connected_nn.train(traind, trainlbl, NN_ARCHITECTURE, N_EPOCHS, LR,128)
endtb = time.time()
endmb = memory_profiler.memory_usage()
train_time_b = endtb - starttb
train_memory_b = endmb[0] - startmb[0]
print("Train set accuracy base: {:2f}".format(accuracy_history_b[-1]))
print("Training time base: {:2f} sec".format(train_time_b))
print("Training memory base: {:2f} mb".format(train_memory_b))
starttesttb = time.time()
starttestmb = memory_profiler.memory_usage()
Y_test_hat_b, _, _ = fully_connected_nn.full_forward_propagation(testd, params_values_b, NN_ARCHITECTURE)
acc_test_b = fully_connected_nn.get_accuracy_value(Y_test_hat_b, testlbl)
pre_test_b, rec_test_b, F1_test_b = fully_connected_nn.get_performance_value(Y_test_hat_b, testlbl)
endttestb = time.time()
endtestmb = memory_profiler.memory_usage()
test_time_b = endttestb - starttesttb
test_memory_b = endtestmb[0] - starttestmb[0]
print("Test set accuracy base: {:2f}".format(acc_test_b))
print("Test set precision base: {:2f}".format(pre_test_b))
print("Test set recall base: {:2f}".format(rec_test_b))
print("Test set score base: {:2f}".format(F1_test_b))
print("Testing time base: {:2f} sec".format(test_time_b))
print("Testing memory base: {:2f} mb".format(test_memory_b))
def advsry(X, Y, epsi):
Y_hat, cashe, _ = fully_connected_nn.full_forward_propagation(X, params_values_b, NN_ARCHITECTURE)
grads_values, grad_w = fully_connected_nn.full_backward_propagation(Y_hat, Y, cashe, params_values_b, NN_ARCHITECTURE)
pert = np.resize(grad_w, X.shape)
pertubated_data = X + np.sign(epsi * pert)
pertubated_data = np.clip(pertubated_data, 0, 1)
return pertubated_data
X_attack = advsry(testd, testlbl, 1.0)
Y_test_hat_b_r, _, _ = fully_connected_nn.full_forward_propagation(X_attack, params_values_b, NN_ARCHITECTURE)
acc_test_b_r = fully_connected_nn.get_accuracy_value(Y_test_hat_b_r, testlbl)
pre_test_b_r, rec_test_b_r, F1_test_b_r = fully_connected_nn.get_performance_value(Y_test_hat_b_r, testlbl)
print("Adversarial set accuracy base: {:2f}".format(acc_test_b_r))
print("Adversarial set precision base: {:2f}".format(pre_test_b_r))
print("Adversarial set recall base: {:2f}".format(rec_test_b_r))
print("Adeverarial set score base: {:2f}".format(F1_test_b_r))
#Y_test_hat_base, _, _ = fully_connected_nn.full_forward_propagation(adv_b, params_values_b, NN_ARCHITECTURE)
#acc_train_base = fully_connected_nn.get_accuracy_value(Y_test_hat_base, trainlbl)
#print("Adversarial train accuracy base: {:2f}".format(acc_train_base))
#epsis = [0, 0.01, 0.1, 0.15]
#acc_testbr = []
#for epsi in epsis:
# X_attack_b = fully_connected_nn.advsry(testd,epsi, grad_a)
# Y_test_hat_b_r, _, _ = fully_connected_nn.full_forward_propagation(X_attack_b, params_values_b, NN_ARCHITECTURE)
# acc_test_b_r = fully_connected_nn.get_accuracy_value(Y_test_hat_b_r, testlbl)
# acc_testbr.append(acc_test_b_r)
# print(f"Epsilon: {epsi}, adversarial accuracy base:{acc_test_b_r:.3f}")
def train_batch(X, Y, nn_architecture, epochs, learning_rate, batch_size, m_batch_size, verbose=False, callback=None):
# initiation of neural net parameters
params_values = fully_connected_nn.init_layers(nn_architecture, 2)
# initiation of regularization parameters
W_tre = 0.01
lambd = 0.01
lambi = 0.01
# initiation of lists storing the history
# of metrics calculated during the learning process
cost_history = []
accuracy_history = []
examples_size = X.shape[1]
batch_number = examples_size // batch_size
m_batch_number = batch_number // m_batch_size
# performing calculations for subsequent iterations
for i in range(epochs):
batch_idx = epochs % batch_number
# Mini-Batch
X_batch = X[:, batch_idx * batch_size : (batch_idx +1) * batch_size]
Y_batch = Y[:, batch_idx * batch_size : (batch_idx +1) * batch_size]
# Micro Batch
batch_midx = epochs % m_batch_number
X_batch = X[:, batch_midx * m_batch_size : (batch_midx +1) * m_batch_size]
Y_batch = Y[:, batch_midx * m_batch_size : (batch_midx +1) * m_batch_size]
# step forward
Y_hat, cashe, W_curr = fully_connected_nn.full_forward_propagation(X_batch, params_values, nn_architecture)
# calculating metrics and saving them in history
cost = fully_connected_nn.compute_cost_with_regularization(Y_hat, Y_batch, eps, W_curr, W_tre, lambd)
cost_history.append(cost)
accuracy = fully_connected_nn.get_accuracy_value(Y_hat, Y_batch)
accuracy_history.append(accuracy)
# step backward - calculating gradient
grads_values, grado = fully_connected_nn.full_backward_propagation(Y_hat, Y_batch, cashe, params_values, nn_architecture)
# updating model state
params_values = fully_connected_nn.update(params_values, grads_values, nn_architecture, learning_rate)
if cost <= cost_history_b[-1]:
lambd = lambd + lambi
if(i % 50 == 0):
if(verbose):
print("Iteration: {:05} - cost: {:.5f} - accuracy: {:.5f}".format(i, cost, accuracy))
if(callback is not None):
callback(i, params_values)
return params_values, cost_history, accuracy_history, grado
starttbatch = time.time()
startmbatch = memory_profiler.memory_usage()
params_values_batch, cost_history_batch, accuracy_history_batch, grado = train_batch(traind, trainlbl, NN_ARCHITECTURE, N_EPOCHS, LR, 128, 4)
endtbatch = time.time()
endmbatch = memory_profiler.memory_usage()
train_time_batch = endtbatch - starttbatch
train_memory_batch = endmbatch[0] - startmbatch[0]
print("Train set accuracy optimize: {:2f}".format(accuracy_history_batch[-1]))
print("Training time optimize: {:2f} sec".format(train_time_batch))
print("Training memory optimize: {:2f} mb".format(train_memory_batch))
starttestbatch = time.time()
starttestmbatch = memory_profiler.memory_usage()
Y_test_hat_batch, _, _ = fully_connected_nn.full_forward_propagation(testd, params_values_batch, NN_ARCHITECTURE)
acc_test_batch = fully_connected_nn.get_accuracy_value(Y_test_hat_batch, testlbl)
pre_test_batch, rec_test_batch, F1_test_batch = fully_connected_nn.get_performance_value(Y_test_hat_batch, testlbl)
endttestbatch = time.time()
endtestmbatch = memory_profiler.memory_usage()
test_time_batch = endttestbatch - starttestbatch
test_memory_batch = endtestmbatch[0] - starttestmbatch[0]
print("Test set accuracy optimize: {:2f}".format(acc_test_batch))
print("Test set precision optimize: {:2f}".format(pre_test_batch))
print("Test set recall optimize: {:2f}".format(rec_test_batch))
print("Test set score optimize: {:2f}".format(F1_test_batch))
print("Testing time optimize: {:2f} sec".format(test_time_batch))
print("Testing memory optimize: {:2f} mb".format(test_memory_batch))
Y_test_hat_batch_r, _, _ = fully_connected_nn.full_forward_propagation(X_attack, params_values_batch, NN_ARCHITECTURE)
acc_test_batch_r = fully_connected_nn.get_accuracy_value(Y_test_hat_batch_r, testlbl)
pre_test_batch_r, rec_test_batch_r, F1_test_batch_r = fully_connected_nn.get_performance_value(Y_test_hat_batch_r, testlbl)
print("Adversarial set accuracy optimize: {:2f}".format(acc_test_batch_r))
print("Adversarial set precision optimize: {:2f}".format(pre_test_batch_r))
print("Adeversarial set recall optimize: {:2f}".format(rec_test_batch_r))
print("Adversarial set score optimize: {:2f}".format(F1_test_batch_r))
#X_attack_o = fully_connected_nn.advsry(testd, 0.15)
#Y_test_hat_batch_r, _, _ = fully_connected_nn.full_forward_propagation(X_attack_o, params_values_batch, NN_ARCHITECTURE)
#acc_test_batch_r = fully_connected_nn.get_accuracy_value(Y_test_hat_batch_r, testlbl)
#print("Adversarial set accuracy optimize: {:2f}".format(acc_test_batch_r))
#acc_testbatchr = []
#for epsi in epsis:
# X_attack_o = fully_connected_nn.advsry(testd, eps, grado)
# Y_test_hat_batch_r, _, _ = fully_connected_nn.full_forward_propagation(X_attack_o, params_values_batch, NN_ARCHITECTURE)
# acc_test_batch_r = fully_connected_nn.get_accuracy_value(Y_test_hat_batch_r, testlbl)
# acc_testbatchr.append(acc_test_batch_r)
# print(f"Epsilon: {epsi}, adversarial accuracy optimize:{acc_test_batch_r:.3f}")
#print("Test adverasarial accuracy optimize: {:2f}".format(acc_test_batch_r))