版权声明:本文为博主原创文章,未经博主允许不得转载。 https://blog.csdn.net/u010916338/article/details/84067604
-
数据集选用KDD99
数据下载地址:http://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html
需求:https://blog.csdn.net/com_stu_zhang/article/details/6987632 -
运行环境
win10+keras
安装步骤:https://blog.csdn.net/u010916338/article/details/83822562 -
数据预处理
包含数值替换文本、数值归一化、标签独热编码
# -*- coding: utf-8 -*-
"""
Created on Tue Nov 6 09:24:20 2018
@author: hrh
"""
import pandas as pd
from sklearn.preprocessing import OneHotEncoder
from pandas.core.frame import DataFrame
def get_total_data():
data = pd.read_csv('data_test.csv',header=None)
data[1]=data[1].map({'tcp':0, 'udp':1, 'icmp':2})
data[2]=data[2].map({'aol':0, 'auth':1, 'bgp':2, 'courier':3, 'csnet_ns':4,'ctf':5, 'daytime':6, 'discard':7, 'domain':8, 'domain_u':9,'echo':10, 'eco_i':11, 'ecr_i':12, 'efs':13, 'exec':14,'finger':15, 'ftp':16, 'ftp_data':17, 'gopher':18, 'harvest':19,'hostnames':20, 'http':21, 'http_2784':22, 'http_443':23, 'http_8001':24,'imap4':25, 'IRC':26, 'iso_tsap':27, 'klogin':28, 'kshell':29,'ldap':30, 'link':31, 'login':32, 'mtp':33, 'name':34,'netbios_dgm':35, 'netbios_ns':36, 'netbios_ssn':37, 'netstat':38, 'nnsp':39,'nntp':40, 'ntp_u':41, 'other':42, 'pm_dump':43, 'pop_2':44,'pop_3':45, 'printer':46, 'private':47, 'red_i':48, 'remote_job':49,'rje':50, 'shell':51, 'smtp':52, 'sql_net':53, 'ssh':54,'sunrpc':55, 'supdup':56, 'systat':57, 'telnet':58, 'tftp_u':59,'tim_i':60, 'time':61, 'urh_i':62, 'urp_i':63, 'uucp':64,'uucp_path':65, 'vmnet':66, 'whois':67, 'X11':68, 'Z39_50':69})
data[3]=data[3].map({'OTH':0, 'REJ':0, 'RSTO':0,'RSTOS0':0, 'RSTR':0, 'S0':0,'S1':0, 'S2':0, 'S3':0,'SF':1, 'SH':0})
data[41]=data[41].map({'normal.':0, 'ipsweep.':1, 'mscan.':2, 'nmap.':3, 'portsweep.':4, 'saint.':5, 'satan.':6, 'apache2.':7,'back.':8, 'land.':9, 'mailbomb.':10, 'neptune.':11, 'pod.':12,'processtable.':13, 'smurf.':14, 'teardrop.':15, 'udpstorm.':16, 'buffer_overflow.':17, 'httptunnel.':18, 'loadmodule.':19, 'perl.':20, 'ps.':21,'rootkit.':22, 'sqlattack.':23, 'xterm.':24, 'ftp_write.':25,'guess_passwd.':26, 'imap.':27, 'multihop.':28, 'named.':29, 'phf.':30,'sendmail.':31, 'snmpgetattack.':32, 'snmpguess.':33, 'spy.':34, 'warezclient.':35,'warezmaster.':36, 'worm.':37, 'xlock.':38, 'xsnoop.':39})
data[2] = (data[2]-data[2].min())/(data[2].max() - data[2].min())
data[4] = (data[4]-data[4].min())/(data[4].max() - data[4].min())
data[5] = (data[5]-data[5].min())/(data[5].max() - data[5].min())
data[22] = (data[22]-data[22].min())/(data[22].max() - data[22].min())
data[23] = (data[23]-data[23].min())/(data[23].max() - data[23].min())
data[31] = (data[31]-data[31].min())/(data[31].max() - data[31].min())
data[32] = (data[32]-data[32].min())/(data[32].max() - data[32].min())
return data
def get_target_data():
data = get_total_data()
enc = OneHotEncoder(sparse = False)
enc.fit([[0], [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]])
result = enc.transform(data[[41]])
return DataFrame(result)
def get_input_data():
data = get_total_data()
del data[41]
return data
if __name__ == '__main__':
data_input = get_input_data()
# data = get_total_data()
data_input.to_csv('data_test_input.csv',header=None,index=None)
data_target = get_target_data()
data_target.to_csv('data_test_target.csv',index=None,header=None)
- 代码
# -*- coding: utf-8 -*-
"""
Created on Mon Nov 5 16:34:42 2018
@author: hrh
"""
import time
start = time.time()
from sklearn.preprocessing import OneHotEncoder
import tensorflow as tf
from keras.models import Sequential
from keras.layers.core import Dense, Activation
import pandas as pd
from keras.optimizers import SGD
if __name__ == '__main__':
input_data = pd.read_csv('data_input.csv',header=None)
target_data = pd.read_csv('data_target.csv',header=None)
input_data_test = pd.read_csv('data_test_input.csv',header=None)
target_data_test = pd.read_csv('data_test_target.csv',header=None)
model = Sequential() #层次模型
model.add(Dense(54, input_dim=41, init='uniform', activation='relu'))
# model.add(Dense(64, input_dim=54, init='uniform', activation='relu'))
model.add(Dense(40, init='uniform', activation='relu'))
model.add(Dense(40, init='uniform', activation='softmax'))
sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
model.fit(input_data, target_data, nb_epoch=12, batch_size=128)
# scores = model.evaluate(input_data, target_data)
# print("%s: %.2f%%" % (model.metrics_names[1], scores[1]*10))
# 将测试集输入到训练好的模型中,查看测试集的误差
loss_and_metrics = model.evaluate(input_data_test, target_data_test, batch_size=128)
result = model.predict(input_data_test, batch_size=128)
print('测试集的预测结果为:', result)
stop = time.time()
print(str(stop-start) + "秒")
- 运行结果
Epoch 1/12
494021/494021 [==============================] - 5s 10us/step - loss: 0.1277 - acc: 0.9720
Epoch 2/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0327 - acc: 0.9946
Epoch 3/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0252 - acc: 0.9954
Epoch 4/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0215 - acc: 0.9963
Epoch 5/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0192 - acc: 0.9967
Epoch 6/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0177 - acc: 0.9971
Epoch 7/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0164 - acc: 0.9973
Epoch 8/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0155 - acc: 0.9976
Epoch 9/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0149 - acc: 0.9977
Epoch 10/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0142 - acc: 0.9978
Epoch 11/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0138 - acc: 0.9979
Epoch 12/12
494021/494021 [==============================] - 4s 9us/step - loss: 0.0134 - acc: 0.9979
311029/311029 [==============================] - 1s 4us/step
Test loss: nan
Test accuracy: 0.9157281153902965
测试集的预测结果为: [[9.9969363e-01 6.1610535e-07 7.2848221e-11 ... 7.0689246e-11
1.6535544e-10 2.0521278e-10]
[9.9969363e-01 6.1610535e-07 7.2848221e-11 ... 7.0689246e-11
1.6535544e-10 2.0521278e-10]
[9.9969363e-01 6.1610535e-07 7.2848221e-11 ... 7.0689246e-11
1.6535544e-10 2.0521278e-10]
...
[9.9966860e-01 6.4044093e-07 7.9545072e-11 ... 7.7094046e-11
1.8038426e-10 2.2377748e-10]
[9.9966013e-01 6.5163550e-07 7.8919801e-11 ... 7.6405368e-11
1.7924477e-10 2.2227653e-10]
[9.9966860e-01 6.4044093e-07 7.9545072e-11 ... 7.7094046e-11
1.8038426e-10 2.2377748e-10]]
61.35442543029785秒