朴素贝叶斯

朴素贝叶斯分类器的构造基础是贝叶斯理论。采用概率模型来表述，定义x=<x1,x2,...,xn>为某一n维特征向量，y∈{c1,c2,...ck}为该特征向量x所有k种可能的类别，记

P(y=ci|x)为特征向量x属于类别ci的概率。贝叶斯原理：

P(y|x)=P(x|y)P(y)/P(x)

#代码1：读取20类新闻文本的数据细节

 #从sklearn.datasets里导入新闻数据抓取器fetch_20newsgroups

from sklearn.datasets import  fetch_20newsgroups

#需要从互联网下载数据

news=fetch_20newsgroups(subset='all')

print(len(news.data))

print(news.data[0])
 
 18846

From: Mamatha Devineni Ratnam <mr47+@Andrew.cmu.edu>

Subject: Pens fans reactions

Organization: Post Office, Carnegie Mellon, Pittsburgh, PA

Lines: 12

NNTP-Posting-Host: po4.andrew.cmu.edu


I am sure some bashers of Pens fans are pretty confused about the lack

of any kind of posts about the recent Pens massacre of the Devils. Actually,

I am  bit puzzled too and a bit relieved. However, I am going to put an end

to non-PIttsburghers' relief with a bit of praise for the Pens. Man, they

are killing those Devils worse than I thought. Jagr just showed you why

he is much better than his regular season stats. He is also a lot

fo fun to watch in the playoffs. Bowman should let JAgr have a lot of

fun in the next couple of games since the Pens are going to beat the pulp out of Jersey anyway. I was very disappointed not to see the Islanders lose the final

regular season game.          PENS RULE!!!
 
可能出现的问题 fetch_20newsgroups 数据集导入失败: 1. 下载20news-bydate.tar.gz（http://people.csail.mit.edu/jrennie/20Newsgroups/20news-bydate.tar.gz）
到C:User\Adminster\scikit_learn_data\20news_home 下

2. 修改 C:\Anaconda\Lib\site-packages\sklearn\datasets\twenty_newsgroups.py 里面的 download_20newsgroups()函数，注释掉下面的代码

  
 # if os.path.exists(archive_path):

    #     # Download is not complete as the .tar.gz file is removed after

    #     # download.

    #     logger.warning("Download was incomplete, downloading again.")

    #     os.remove(archive_path)


    # logger.warning("Downloading dataset from %s (14 MB)", URL)


    # opener = urlopen(URL)

    # with open(archive_path, 'wb') as f:

    #     f.write(opener.read())

并修改

    archive_path = os.path.join(target_dir, r'20newsbydate.tar.gz')

3. 运行, fetch_20newsgroups会自动解压20news-bydate.tar.gz，生成缓存文件20news-bydate_py3.pkz路径为(C:User\Adminster\scikit_learn_data\20news-bydate_py3.pkz)
 
 #20类新闻文本数据分割

from sklearn.cross_validation import train_test_split

#随机采样25%的数据用于测试，剩下的75%用于构建训练集合

X_train,X_test,y_train,y_test=train_test_split(news.data,news.target,test_size=0.25,random_state=33)

#使用朴素贝叶斯分类器对新闻文本数据进行类别预测

#从sklearn.feature_extraction.text里导入用于文本特征向量转换模块

from sklearn.feature_extraction.text import CountVectorizer

vec=CountVectorizer()

X_train=vec.fit_transform(X_train)

X_test=vec.transform(X_test)

#从sklearn.naive_bayas里导入朴素贝叶斯模型

from sklearn.naive_bayes import MultinomialNB

mnb=MultinomialNB()

#利用训练数据对模型参数进行估计

mnb.fit(X_train,y_train)

#预测结果存储在变量y_predict中

y_predict=mnb.predict(X_test)

#对朴素贝叶斯分类器在新闻文本数据上的表现性能进行评估

#使用模型自带的评估函数进行准确性测评

print('The Accuracy of Naïve Bayes is',mnb.score(X_test,y_test))

#从sklearn.metrics里导入classification_report模块

from sklearn.metrics import classification_report

print(classification_report(y_test,y_predict,target_names=news.target_names))
 
 
 
 The Accuracy of Naïve Bayes is 0.8397707979626485

                          precision    recall  f1-score   support



             alt.atheism       0.86      0.86      0.86       201

           comp.graphics       0.59      0.86      0.70       250

 comp.os.ms-windows.misc       0.89      0.10      0.17       248

comp.sys.ibm.pc.hardware       0.60      0.88      0.72       240

   comp.sys.mac.hardware       0.93      0.78      0.85       242

          comp.windows.x       0.82      0.84      0.83       263

            misc.forsale       0.91      0.70      0.79       257

               rec.autos       0.89      0.89      0.89       238

         rec.motorcycles       0.98      0.92      0.95       276

      rec.sport.baseball       0.98      0.91      0.95       251

        rec.sport.hockey       0.93      0.99      0.96       233

               sci.crypt       0.86      0.98      0.91       238

         sci.electronics       0.85      0.88      0.86       249

                 sci.med       0.92      0.94      0.93       245

               sci.space       0.89      0.96      0.92       221

  soc.religion.christian       0.78      0.96      0.86       232

      talk.politics.guns       0.88      0.96      0.92       251

   talk.politics.mideast       0.90      0.98      0.94       231

      talk.politics.misc       0.79      0.89      0.84       188

      talk.religion.misc       0.93      0.44      0.60       158



             avg / total       0.86      0.84      0.82      4712