-
Notifications
You must be signed in to change notification settings - Fork 0
/
利用keras对kdd99数据进行分类.py
143 lines (105 loc) · 5.11 KB
/
利用keras对kdd99数据进行分类.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
import pandas as pd
from sklearn.model_selection import train_test_split
from keras.layers import Dense
from keras.utils import np_utils
from keras.models import Sequential
from keras.optimizers import Adam
import keras as K
from sklearn.preprocessing import MinMaxScaler
import tensorflow as tf
import keras.losses as K_loss
'''
分别使用了标准的交叉熵和防止过拟合化的均匀分布的交叉熵和svm_loss
'''
data = pd.read_csv('./kddcup99_10%.csv', engine='python')
columns_list = ['duration', 'protocol_type', 'service', 'flag', 'src_bytes', 'dst_bytes', 'land', 'wrong_fragment',
'urgent',
'hot', 'num_failed_logins', 'logged_in', 'num_compromised', 'root_shell', 'su_attempted', 'num_root',
'num_file_creations', 'num_shells', 'num_access_files', 'num_outbound_cmds', 'is_hot_login',
'is_guest_login',
'count', 'srv_count', 'serror_rate', 'srv_serror_rate', 'rerror_rate', 'srv_rerror_rate',
'same_srv_rate', 'diff_srv_rate', 'srv_diff_host_rate',
'dst_host_count', 'dst_host_srv_count', 'dst_host_same_srv_rate', 'dst_host_diff_srv_rate',
'dst_host_same_src_port_rate', 'dst_host_srv_diff_host_rate', 'dst_host_serror_rate',
'dst_host_srv_serror_rate', 'dst_host_rerror_rate', 'dst_host_srv_rerror_rate',
'label']
x_lists = ['duration', 'protocol_type', 'service', 'flag', 'src_bytes', 'dst_bytes', 'land', 'wrong_fragment',
'urgent',
'hot', 'num_failed_logins', 'logged_in', 'num_compromised', 'root_shell', 'su_attempted', 'num_root',
'num_file_creations', 'num_shells', 'num_access_files', 'num_outbound_cmds', 'is_hot_login',
'is_guest_login',
'count', 'srv_count', 'serror_rate', 'srv_serror_rate', 'rerror_rate', 'srv_rerror_rate',
'same_srv_rate', 'diff_srv_rate', 'srv_diff_host_rate',
'dst_host_count', 'dst_host_srv_count', 'dst_host_same_srv_rate', 'dst_host_diff_srv_rate',
'dst_host_same_src_port_rate', 'dst_host_srv_diff_host_rate', 'dst_host_serror_rate',
'dst_host_srv_serror_rate', 'dst_host_rerror_rate', 'dst_host_srv_rerror_rate',]
data_x, data_y = data[x_lists], data['label']
# print(train_x.shape) # (444618, 41)
# print(test_x.shape) # (49403, 41)
minmax = MinMaxScaler()
train_x = minmax.fit_transform(data_x)
one_hot_train_y = np_utils.to_categorical(data_y, num_classes=23)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> softmax-categorical_crossentropy<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
# 经过均匀分布的交叉熵
# 这玩意具体意思就是说90%利用正确数据标签[1,0·····0]的交叉熵，10%用[1/23,1/23···1/23]共23个均匀分布的交叉熵
# def my_categorical_crossentropy(y_true, y_pred, e=0.1):
# loss1 = K_loss.categorical_crossentropy(y_true, y_pred)
# loss2 = K_loss.categorical_crossentropy(K.backend.ones_like(y_pred)/23, y_pred)
# return (1-e) * loss1 + e * loss2
def svm_loss(y_true, y_pred, delta=1.):
yi = K.backend.max(y_true * y_pred, axis=-1)
yi = K.backend.reshape(yi, (-1, 1))
one_ = y_pred - yi + delta
losses = K.backend.sum(K.backend.maximum(one_ - y_true, 0.), axis=-1)
return K.backend.mean(losses)
def linear_regression_equality(y_true, y_pred):
diff = K.backend.abs(y_true-y_pred)
return K.backend.mean(K.backend.cast(diff < 0.001, tf.float32))
model = Sequential()
# 41->64->128->64->32->23
model.add(Dense(64, input_dim=41, activation='relu'))
model.add(Dense(128, activation='relu'))
model.add(Dense(64, activation='relu'))
model.add(Dense(32, activation='relu'))
model.add(Dense(23, activation='softmax')) # 多分类问题用概率来
adam = Adam(0.001)
# 普通交叉熵， 过份自信在训练样本会出现非常高的准确率，导致在测试期间会出现过拟合的现象
# model.compile(
# optimizer=adam,
# loss="categorical_crossentropy",
# metrics=["accuracy"]
# )
# 均匀分布过的交叉熵：
# model.compile(
# optimizer=adam,
# loss=my_categorical_crossentropy,
# metrics=["accuracy"]
# )
# svm_loss
model.compile(
optimizer=adam,
loss=svm_loss,
metrics=["accuracy"]
)
# model.fit(train_x, one_hot_train_y, epochs=10, batch_size=1000, shuffle=True)
# 对于均匀交叉熵，这里10轮和20轮会出现不同的准确率，
model.fit(train_x, one_hot_train_y, epochs=10, batch_size=10000, shuffle=True)
pred = model.predict(train_x)
print(pred[0])
print(data_y[0])
# predict = model.predict(test_x)
# print(test_x.values[:1, :])
# print(predict[0])
# loss = model.evaluate(test_x, one_hot_test_y)
# print(loss)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>softmax-categorical_crossentropy<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
# 定义多分类svm_loss
# def svm_loss(y_true, y_pred, margin=1, train_batch=1000):
# loss_list = []
# for i in y_pred:
# one_loss = 0
# for j in i:
# if j != y_true:
# one_loss += max(i[j] - i[y_true])
# loss_list.append(one_loss)
# return sum(loss_list)/train_batch