朴素贝叶斯分类算法 & sklearn中的朴素贝叶斯模型及其应用 & 朴素贝叶斯应用:垃圾邮件分类

简述分类与聚类的联系与区别。

  分类是指在对数据集分类时,我们知道这个数据集是有多少种类的。

  聚类是将数据对象的集合分成相似的对象类的过程,使得同一个簇(或类)中的对象之间具有较高的相似性,而不同簇中的对象具有较高的相异性。即指在对数据集操作时,我们是不知道该数据集包含多少类,我们要做的,是将数据集中相似的数据归纳在一起。

简述什么是监督学习与无监督学习。

  监督学习是指每个实例都是由一组特征和一个类别结果,拥有标注的数据训练模型,并产生一个推断的功能。对于新的实例,可以用于映射出该实例的类别。

  无监督学习是指我们只知道一些特征,并不知道答案,但不同实例具有一定的相似性,然后把那些相似的聚集在一起。

2.朴素贝叶斯分类算法 实例

利用关于心脏情患者的临床数据集,建立朴素贝叶斯分类模型。

有六个分类变量(分类因子):性别,年龄、KILLP评分、饮酒、吸烟、住院天数

目标分类变量疾病:–心梗–不稳定性心绞痛

新的实例:–(性别=‘男’,年龄<70, KILLP=‘I',饮酒=‘是’,吸烟≈‘是”,住院天数<7)

最可能是哪个疾病?

 

3、编程实现朴素贝叶斯分类算法

import pandas as pd
import numpy as np
 
dataDF = pd.read_excel(r'data/心脏病患者临床数据.xlsx')
 
# 数据处理,对男女(男1女0),年龄(<70 -1,70-80 0,>80 1),
# 住院天数(<7 -1,7-14 0,>14 1)三个列进行处理
sex = []
for s in dataDF['性别']:
    if s == '男':
        sex.append(1)
    else:
        sex.append(0)
 
age = []
for a in dataDF['年龄']:
    if a == '<70':
        age.append(-1)
    elif a == '70-80':
        age.append(0)
    else:
        age.append(1)
 
days = []
for d in dataDF['住院天数']:
    if d == '<7':
        days.append(-1)
    elif d == '7-14':
        days.append(0)
    else:
        days.append(1)
 
# 另外生成一份处理后的DF
dataDF2 = dataDF
dataDF2['性别'] = sex
dataDF2['年龄'] = age
dataDF2['住院天数'] = days
 
# 转为数组用于计算
dataarr = np.array(dataDF)
dataarr
 
# 用贝叶斯模型判断病人属于哪种病:性别=‘男’,年龄<70, KILLP=1,饮酒=‘是’,吸烟=‘是”,住院天数<7
def beiyesi(sex, age, KILLP, drink, smoke, days):
    # 初始化变量
    x1_y1,x2_y1,x3_y1,x4_y1,x5_y1,x6_y1 = 0,0,0,0,0,0
    x1_y2,x2_y2,x3_y2,x4_y2,x5_y2,x6_y2 = 0,0,0,0,0,0
    y1 = 0
    y2 = 0
     
    for line in dataarr:
        if line[6] == '心梗':# 计算在心梗条件下出现各症状的次数
            y1 += 1
            if line[0] == sex:
                x1_y1 += 1
            if line[1] == age:
                x2_y1 += 1
            if line[2] == KILLP:
                x3_y1 += 1
            if line[3] == drink:
                x4_y1 += 1
            if line[4] == smoke:
                x5_y1 += 1
            if line[5] == days:
                x6_y1 += 1
        else: # 计算不稳定性心绞痛条件下出现各症状的次数
            y2 += 1
            if line[0] == sex:
                x1_y2 += 1
            if line[1] == age:
                x2_y2 += 1
            if line[2] == KILLP:
                x3_y2 += 1
            if line[3] == drink:
                x4_y2 += 1
            if line[4] == smoke:
                x5_y2 += 1
            if line[5] == days:
                x6_y2 += 1
    # print('y1:',y1,' y2:',y2)
             
             
    # 计算,转为x|y1, x|y2
    # print('x1_y1:',x1_y1, ' x2_y1:',x2_y1, ' x3_y1:',x3_y1, ' x4_y1:',x4_y1, ' x5_y1:',x5_y1, ' x6_y1:',x6_y1)
    # print('x1_y2:',x1_y2, ' x2_y2:',x2_y2, ' x3_y2:',x3_y2, ' x4_y2:',x4_y2, ' x5_y2:',x5_y2, ' x6_y2:',x6_y2)
    x1_y1, x2_y1, x3_y1, x4_y1, x5_y1, x6_y1 = x1_y1/y1, x2_y1/y1, x3_y1/y1, x4_y1/y1, x5_y1/y1, x6_y1/y1
    x1_y2, x2_y2, x3_y2, x4_y2, x5_y2, x6_y2 = x1_y2/y2, x2_y2/y2, x3_y2/y2, x4_y2/y2, x5_y2/y2, x6_y2/y2
    x_y1 = x1_y1 * x2_y1 * x3_y1 * x4_y1 * x5_y1 * x6_y1
    x_y2 = x1_y2 *  x2_y2 * x3_y2 * x4_y2 * x5_y2 * x6_y2
 
         
    # 计算各症状出现的概率
    x1,x2,x3,x4,x5,x6 = 0,0,0,0,0,0
    for line in dataarr:
        if line[0] == sex:
            x1 += 1
        if line[1] == age:
            x2 += 1
        if line[2] == KILLP:
            x3 += 1
        if line[3] == drink:
            x4 += 1
        if line[4] == smoke:
            x5 += 1
        if line[5] == days:
            x6 += 1
    # print('x1:',x1, ' x2:',x2, ' x3:',x3, ' x4:',x4, ' x5:',x5, ' x6:',x6)
    # 计算
    length = len(dataarr)
    x = x1/length * x2/length * x3/length * x4/length * x5/length * x6/length
    # print('x:',x)
     
    # 分别计算 给定症状下心梗 和 不稳定性心绞痛 的概率
    y1_x = (x_y1)*(y1/length)/x
    # print(y1_x)
    y2_x = (x_y2)*(y2/length)/x
     
    # 判断是哪中疾病的可能性大
    if y1_x > y2_x:
        print('该病人患心梗的可能性较大,可能性为:',y1_x)
    else:
        print('该病人患不稳定性心绞痛的可能性较大,可能性为:',y2_x)
 
# 判断:性别=‘男’,年龄<70, KILLP=1,饮酒=‘是’,吸烟=‘是”,住院天数<7
beiyesi(1,-1,1,'是','是',-1)

 

结果为:

 

 

1.使用朴素贝叶斯模型对iris数据集进行花分类

尝试使用3种不同类型的朴素贝叶斯:

高斯分布型

from sklearn import datasets
iris = datasets.load_iris()

from sklearn.naive_bayes import GaussianNB
Gaus = GaussianNB()
pred = Gaus.fit(iris.data , iris.target)
G_pred = pred.predict(iris.data)

print(iris.data.shape[0],(iris.target !=G_pred).sum())

print(iris.target)

 

伯努利型

from sklearn.naive_bayes import BernoulliNB
from sklearn import datasets
iris = datasets.load_iris()
Bern = BernoulliNB()
pred = Bern.fit(iris.data, iris.target)
B_pred = pred.predict(iris.data)
print(iris.data.shape[0],(iris.target !=B_pred).sum())
print(iris.target)
print(B_pred)

 

多项式型

from sklearn import datasets
from sklearn.naive_bayes import MultinomialNB
iris = datasets.load_iris()
Mult = MultinomialNB()
pred = Mult.fit(iris.data, iris.target)
M_pred = pred.predict(iris.data)
print(iris.data.shape[0],(iris.target !=M_pred).sum())
print(iris.target)
print(M_pred)
print(iris.target !=M_pred)

 

2.使用sklearn.model_selection.cross_val_score(),对模型进行验证。

 

from sklearn.naive_bayes import GaussianNB
from sklearn.model_selection import cross_val_score
gnb = GaussianNB()
scores = cross_val_score(gnb,iris.data,iris.target,cv=10)
print("准确率:%.3f"%scores.mean())

from sklearn.naive_bayes import BernoulliNB
from sklearn.model_selection import cross_val_score
Bern = BernoulliNB()
scores = cross_val_score(Bern,iris.data,iris.target,cv=10)
print("准确率:%.3f"%scores.mean())

from sklearn.naive_bayes import MultinomialNB
from sklearn.model_selection import cross_val_score
Mult = MultinomialNB()
scores = cross_val_score(Mult,iris.data,iris.target,cv=10)
print("准确率:%.3f"%scores.mean())

 

 

3. 垃圾邮件分类

数据准备:

  • 用csv读取邮件数据,分解出邮件类别及邮件内容。
  • 对邮件内容进行预处理:去掉长度小于3的词,去掉没有语义的词等

(1)读取数据集

import csv
file_path = r'D:\shuju\SMSSpamCollectionjsn.txt'
mail = open(file_path,'r',encoding='utf-8')
mail_data=[]
mail_label=[]
csv_reader = csv.reader(mail,delimiter='\t')
for line in csv_reader:
    mail_data.append(line[1])
    mail_label.append(line[0])
mail.close()
mail_data

  

 

(2)邮件预处理

 

import nltk
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
from sklearn.naive_bayes import MultinomialNB as MNB  

def preprocessing(text):    
    #text=text.decode("utf-8)
    tokens=[word for sent in nltk.sent_tokenize(text) for word in nltk.word_tokenize(sent)] #nltk进行分词  
    stops=stopwords.words('english')                                  #去掉停用词
    tokens=[token for token in tokens if token not in stops]
    
    tokens=[token.lower() for token in tokens if len(token)>=3]                  #去掉大小写
    lmtzr=WordNetLemmatizer()                                      #词性还原
    tokens=[lmtzr.lemmatize(token) for token in tokens]
    preprocessed_text=' '.join(tokens)
    return preprocessed_text

 

(3)导入数据

import csv
file_path=r'H:\作业\py数据挖掘基础算法\2018.12.3\SMSSpamCollectionjsn.txt'
sms=open(file_path,'r',encoding='utf-8')
sms_data=[]
sms_label=[]
csv_reader=csv.reader(sms,delimiter='\t')
for line in csv_reader:
    sms_label.append(line[0])
    sms_data.append(preprocessing(line[1]))
sms.close()

  

(4)训练集与测试集:将先验数据按一定比例进行拆分。

import numpy as np
sms_data=np.array(sms_data)
sms_label=np.array(sms_label)

from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test =train_test_split(sms_data, sms_label, test_size=0.3, random_state=0, stratify=sms_label) #按训练集和测试集0.7:0.3划分
print(len(sms_data),len(x_train),len(x_test))
x_train

 

(5)提取数据特征,将文本解析为词向量 。

from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer=TfidfVectorizer(min_df=2,ngram_range=(1,2),stop_words='english',strip_accents='unicode',norm='l2')
X_train=vectorizer.fit_transform(x_train)
X_test=vectorizer.transform(x_test)

 

(6)训练模型:建立模型,用训练数据训练模型。即根据训练样本集,计算词项出现的概率P(xi|y),后得到各类下词汇出现概率的向量 。

X_train
a=X_train.toarray()
print(a)

for i in range(1000):
    for j in range(5984):
        if a[i,j]!=0:
            print(i,j,a[i,j])

 

(7)

测试模型:用测试数据集评估模型预测的正确率。

混淆矩阵

准确率、精确率、召回率、F值

from sklearn.naive_bayes import MultinomialNB as MNB
from sklearn.metrics import confusion_matrix
from sklearn.metrics import classification_report

clf=MNB().fit(X_train,y_train)
y_nb_pred=clf.predict(X_test)

print(y_nb_pred.shape, y_nb_pred)
print('nb_confusion_matrix:')              #混淆矩阵
cm = confusion_matrix(y_test, y_nb_pred)
print(cm)                          #准确率、精确率、召回率、F值
cr = classification_report(y_test, y_nb_pred)
print(cr)

feature_names = vectorizer.get_feature_names()
coefs = clf.coef_
intercept = clf.intercept_
coefs_with_fns = sorted(zip(coefs[0], feature_names))

n = 10
top = zip(coefs_with_fns[:n], coefs_with_fns[:-(n + 1):-1])  #
for (coef_1, fn_1), (coef_2, fn_2) in top:
    print('\t%.4f\t%-15s\t\t%.4f\t%-15s' % (coef_1, fn_1, coef_2, fn_2))

 

 

 

 

posted @ 2018-11-26 09:03  崔泽贤  阅读(798)  评论(0编辑  收藏