《机器学习实战（Scala实现）》（四）——朴素贝叶斯

原理

关于算法原理可以参阅：http://blog.csdn.net/u011239443/article/details/53735609#t35

构建词向量

python

def loadDataSet():
postingList=[['my', 'dog', 'has', 'flea', 'problems', 'help', 'please'],
['maybe', 'not', 'take', 'him', 'to', 'dog', 'park', 'stupid'],
['my', 'dalmation', 'is', 'so', 'cute', 'I', 'love', 'him'],
['stop', 'posting', 'stupid', 'worthless', 'garbage'],
['mr', 'licks', 'ate', 'my', 'steak', 'how', 'to', 'stop', 'him'],
['quit', 'buying', 'worthless', 'dog', 'food', 'stupid']]
# 1 代表侮辱性的词 0则不是
classVec = [0,1,0,1,0,1]
return postingList,classVec

def createVocabList(dataSet):
# 创建空集合
vocabSet = set([])
for document in dataSet:
# 合并两个集合
vocabSet = vocabSet | set(document)
return list(vocabSet)

def setOfWords2Vec(vocabList, inputSet):
returnVec = [0]*len(vocabList)
for word in inputSet:
if word in vocabList:
returnVec[vocabList.index(word)] = 1
else: print "the word: %s is not in my Vocabulary!" % word
return returnVec

训练与测试算法

python

训练算法

def trainNB0(trainMatrix,trainCategory):
numTrainDocs = len(trainMatrix)
numWords = len(trainMatrix[0])
pAbusive = sum(trainCategory)/float(numTrainDocs)
p0Num = ones(numWords); p1Num = ones(numWords)
p0Denom = 2.0; p1Denom = 2.0
for i in range(numTrainDocs):
if trainCategory[i] == 1:
p1Num += trainMatrix[i]
p1Denom += sum(trainMatrix[i])
else:
p0Num += trainMatrix[i]
p0Denom += sum(trainMatrix[i])
p1Vect = log(p1Num/p1Denom)
p0Vect = log(p0Num/p0Denom)
return p0Vect,p1Vect,pAbusive

这里的

pAbusive

其实应该计算的是各个类别的概率。但是我们这里是类别只有 0 和 1 的二分类，所以只要返回一个类别为 1 的概率给后续程序就行了。

p0Num = ones(numWords); p1Num = ones(numWords)

和

p0Denom = 2.0; p1Denom = 2.0

是为了避免后续计算log中的指和分母值取到

0

。

p1Num/p1Denom

得到向量第i个特征即p(wi/c1)，而log(p(w/c1))=log(p(w1/c1))+log(p(w2/c1))+...+log(p(wn/c1))。所以我们只需要将该向量中的每个特征取对数，再累加就能得到log(p(w/c1))

测试算法

def classifyNB(vec2Classify, p0Vec, p1Vec, pClass1):
p1 = sum(vec2Classify * p1Vec) + log(pClass1)
p0 = sum(vec2Classify * p0Vec) + log(1.0 - pClass1)
if p1 > p0:
return 1
else:
return 0

def testingNB():
listOPosts,listClasses = loadDataSet()
myVocabList = createVocabList(listOPosts)
trainMat=[]
for postinDoc in listOPosts:
trainMat.append(setOfWords2Vec(myVocabList, postinDoc))
p0V,p1V,pAb = trainNB0(array(trainMat),array(listClasses))
testEntry = ['love', 'my', 'dalmation']
thisDoc = array(setOfWords2Vec(myVocabList, testEntry))
print testEntry,'classified as: ',classifyNB(thisDoc,p0V,p1V,pAb)
testEntry = ['stupid', 'garbage']
thisDoc = array(setOfWords2Vec(myVocabList, testEntry))
print testEntry,'classified as: ',classifyNB(thisDoc,p0V,p1V,pAb)

由于p(w)是相同的，所以我们只需要比较p(w/c0)p(c0)和p(w/c1)p(c1)的大小,即log(p(w/c0)p(c0))和log(p(w/c1)p(c1))的大小。如：log(p(w/c0)p(c0))=log(p(w/c0))+log(p(c0))=

sum(vec2Classify * p0Vec) + log(1.0 - pClass1)

scala

package NativeBayes

import scala.collection.mutable.ArrayBuffer

object NativeBayes {

def loadDataSet() = {
val postingList = Array(Array("my", "dog", "has", "flea", "problems", "help", "please"),
Array("maybe", "not", "take", "him", "to", "dog", "park", "stupid"),
Array("my", "dalmation", "is", "so", "cute", "I", "love", "him"),
Array("stop", "posting", "stupid", "worthless", "garbage"),
Array("mr", "licks", "ate", "my", "steak", "how", "to", "stop", "him"),
Array("quit", "buying", "worthless", "dog", "food", "stupid"))
//1 代表不良信息, 反之为 0
val classVec = Array(0, 1, 0, 1, 0, 1)
(postingList, classVec)
}

def setOfWords2Vec(vocabList: Array[String], inputSet: Array[String]) = {
val returnVec = new Array[Int](vocabList.length)
val vocabListWithIndex = vocabList.zipWithIndex
for (word <- inputSet) {
if (vocabList.contains(word))
returnVec(vocabListWithIndex.filter(_._1 == word)(0)._2) = 1
else printf("the word: %s is not in my Vocabulary!\n", word)
}
returnVec
}

def trainNB0(trainMatrix: Array[Array[Int]], trainCategory: Array[Int]) = {
val numTrainDocs = trainMatrix.length
val numWords = trainMatrix(0).length
val pAbusive = trainCategory.sum / numTrainDocs.toDouble
var p0Num = Array.fill(numWords)(1)
var p1Num = Array.fill(numWords)(1)
var p0Denom = 2.0
var p1Denom = 2.0
for (i <- 0 to numTrainDocs - 1) {
if (trainCategory(i) == 1) {
var cnt = 0
p1Num = p1Num.map { x =>
val v = x + trainMatrix(i)(cnt)
cnt += 1
v
}
p1Denom += trainMatrix(i).sum
} else {
var cnt = 0
p0Num = p0Num.map { x =>
val v = x + trainMatrix(i)(cnt)
cnt += 1
v
}
p0Denom += trainMatrix(i).sum
}
}
(p1Num.map(x => math.log(x / p1Denom)), p0Num.map(x => Math.log(x / p0Denom)), pAbusive)
}

def classifyNB(vec2Classify: Array[Int], p0Vec: Array[Double], p1Vec: Array[Double], pClass1: Double) = {
var cnt = 0
val p1 = vec2Classify.map { x =>
val v = x * p1Vec(cnt)
cnt += 1
v
}.sum + math.log(pClass1)
cnt = 0
val p0 = vec2Classify.map { x =>
val v = x * p0Vec(cnt)
cnt += 1
v
}.sum + math.log(1.0 - pClass1)

if (p1 > p0) 1 else 0
}

def main(args: Array[String]): Unit = {
val DataSet = loadDataSet()
val listOPosts = DataSet._1
val listClasses = DataSet._2
val myVocabList = listOPosts.reduce((a1, a2) => a1.++:(a2)).distinct
var trainMat = new ArrayBuffer[Array[Int]](listOPosts.length)
listOPosts.foreach(postinDoc => trainMat.append(setOfWords2Vec(myVocabList, postinDoc)))

val p = trainNB0(trainMat.toArray, listClasses)
val p0V = p._2
val p1V = p._1
val pAb = p._3
val testEntry = Array("love", "my", "dalmation")
val thisDoc = setOfWords2Vec(myVocabList, testEntry)
println(testEntry.mkString(",") + " classified as: " + classifyNB(thisDoc, p0V, p1V, pAb))
val testEntry2 = Array("stupid", "garbage")
val thisDoc2 = setOfWords2Vec(myVocabList, testEntry2)
println(testEntry2.mkString(",") + " classified as: " + classifyNB(thisDoc2, p0V, p1V, pAb))
}
}