训练集链接
提取码:axpf
训练集(正常邮件)结构截图:
训练集里面正常邮件normal和垃圾邮件spam各有24封,利用这些数据训练出模型并对两份待分类邮件进行分类。
邮件长这样:
关于如何利用朴素贝叶斯进行分类,请参考:朴素贝叶斯(Naive Bayes)原理+编程实现拉普拉斯修正的朴素贝叶斯分类器
分类实现过程:
- 首先需要对每一封邮件进行切割处理,得到包含所有词语的列表,处理方法请参考:Python读取有空行的txt文件+将内容分割保存到列表中
- 训练模型,利用贝叶斯公式计算出后验概率
- 得到结果
完整代码:
# -*- coding: utf-8 -*-
"""
@Time : 2020/12/19 19:48
@Author :KI
@File :NaiveBayes_email.py
@Motto:Hungry And Humble
"""
# 读取所有训练数据并按照空格分隔,保存在一个列表里返回
def load_file(path):
cab = []
for i in range(1, 25):
data = open(path % i)
for line in data.readlines():
cab.append(line.strip().split(','))
cab_f = []
for i in range(len(cab)):
for j in range(len(cab[i])):
if cab[i][j] != '':
cab_f.append(cab[i][j].strip())
cab_final = []
for i in cab_f:
for j in i.split(' '):
cab_final.append(j)
return cab_final
# 朴素贝叶斯分类器
def bayes(sample):
path1 = 'Emails/Training/normal/%d.txt'
path2 = 'Emails/Training/spam/%d.txt'
normal_data = load_file(path1)
spam_data = load_file(path2)
# 计算p(x|C1)=p1与p(x|C2)=p2
p1 = 1.0
p2 = 1.0
for i in range(len(sample)):
x = 0.0
for j in normal_data:
if sample[i] == j:
x = x + 1.0
p1 = p1 * ((x + 1.0) / (len(normal_data) + 2.0)) # 拉普拉斯平滑
for i in range(len(sample)):
x = 0.0
for j in spam_data:
if sample[i] == j:
x = x + 1.0
p2 = p2 * ((x + 1.0) / (len(spam_data) + 2.0)) # 拉普拉斯平滑
pc1 = len(normal_data) / (len(normal_data) + len(spam_data))
pc2 = 1 - pc1
if p1 * pc1 > p2 * pc2:
return 'normal'
else:
return 'spam'
# 测试
def test(path):
data = open(path)
cab = []
for line in data.readlines():
cab.append(line.strip().split(','))
cab_f = []
for i in range(len(cab)):
for j in range(len(cab[i])):
if cab[i][j] != '':
cab_f.append(cab[i][j].strip())
cab_final = []
for i in cab_f:
for j in i.split(' '):
cab_final.append(j)
return bayes(cab_final)
if __name__ == '__main__':
print(test('Emails/test/normal.txt'))
print(test('Emails/test/spam.txt'))
sum1 = 0
sum2 = 0
# 再试试训练集
for i in range(1, 25):
if test('Emails/Training/normal/%d.txt' % i) == 'normal':
sum1 = sum1 + 1
for i in range(1, 25):
if test('Emails/Training/spam/%d.txt' % i) == 'spam':
sum2 = sum2 + 1
print('normal分类正确率:', sum1 / 24)
print('spam分类正确率:', sum2 / 24)
运行结果:
normal
spam
normal分类正确率: 0.9583333333333334
spam分类正确率: 1.0
可以看到,分类效果还不错!!