機器學習實戰(Machine Learning in Action)學習筆記————04.樸素貝葉斯分類(bayes)
機器學習實戰(Machine Learning in Action)學習筆記————04.樸素貝葉斯分類(bayes)
關鍵字:樸素貝葉斯、python、原始碼解析
作者:米倉山下
時間:2018-10-25
機器學習實戰(Machine Learning in Action,@author: Peter Harrington)
原始碼下載地址:https://www.manning.com/books/machine-learning-in-action
[email protected]:pbharrin/machinelearninginaction.git
*************************************************************
一、朴樹貝葉斯分類(bayes)
#樸素貝葉斯實現文字分類原理:
bayes公式:P(ci|w)=P(w|ci)P(ci)/p(w)
首先計算侮辱性言論和非侮辱性言論文件出現的概率P(ci)即P(1)、P(0)的概率;接著計算P(w|ci),假設詞的出現相互獨立(也是樸素貝葉斯“樸素”的含義),p(w0,w1,…wN|c1)=p(w0|c1)p(w1|c1)…p(wN|c1),然後利用公式計算在該詞向量下屬於不同類別概率,比較返回概率最大的類別
訓練:P(w|ci)——分類ci下詞向量w的條件概率,P(w|ci)=p(w0,w1,…wN|c1)=p(w0|c1)p(w1|c1)…p(wN|c1);P(ci)——分類ci出現的概率;p(w)——詞向量w出現的概率(每個詞出現的概率之積,是一個定值?)求解:P(ci|w)——詞向量w下分類為ci的條件概率
測試:利用公式計算在該詞向量下屬於不同類別概率,比較返回概率最大的類別
--------------------------------------------------------------
#樸素貝葉斯實現文字分類,訓練函式
#input:trainMatrix——訓練樣本(詞集模型,詞出現與否分別為1和0),trainCategory——類別向量
#output:p0Vect——[2],p1Vect——[1],pAbusive——樣本為侮辱性P(1)的概率
def trainNB0(trainMatrix,trainCategory): numTrainDocs = len(trainMatrix) #訓練樣本數目 numWords = len(trainMatrix[0]) #特徵個數(詞彙表大小) pAbusive = sum(trainCategory)/float(numTrainDocs) #樣本為侮辱性P(1)的概率,即P(ci) p0Num = ones(numWords); p1Num = ones(numWords) #初始化概率,[注1] p0Denom = 2.0; p1Denom = 2.0 for i in range(numTrainDocs): #遍歷樣本 if trainCategory[i] == 1: #侮辱性樣本 p1Num += trainMatrix[i] #侮辱性詞向量之和 p1Denom += sum(trainMatrix[i]) #侮辱性樣本中,出現的詞彙總數else: #非侮辱性樣本 p0Num += trainMatrix[i] #向量之和 p0Denom += sum(trainMatrix[i])#非侮辱性樣本,出現的詞彙總數 p1Vect = log(p1Num/p1Denom) #[1]侮辱性樣本中,詞彙表每個詞彙出現概率,即P(w|ci),[注1] p0Vect = log(p0Num/p0Denom) #[2]非侮辱性樣本中,詞彙表每個詞彙出現概率,同為P(w|ci) return p0Vect,p1Vect,pAbusive
#注:利用貝葉斯分類器對文件進行分類時,要計算多個概率的乘積以獲取文件屬於某個類別的概率(p(w0|c=1)p(w1|c=1)p(w2|c=1)),若其中一個概率為零則整體為零,為避免影響,將p0Num,p1Num初始化為1,將分母p0Denom初始化為2;另外為避免太多小數相乘造成資料下溢,因此對其取對數(f(x)與ln(f(x))具有相同的單調性),並有ln(a*b)=ln(a)+n(b),將其後的乘積運算轉換為加和
#樸素貝葉斯分類測試函式
def classifyNB(vec2Classify, p0Vec, p1Vec, pClass1): #計算屬於不同類別的概率並進行比較 p1 = sum(vec2Classify * p1Vec) + log(pClass1) #計算概率p(w0|c1)p(w1|c1)…p(wN|c1)p(c1), p0 = sum(vec2Classify * p0Vec) + log(1.0 - pClass1) #因為進行了對數運算,乘積運算轉換為加和 if p1 > p0: return 1 else: return 0
-------------------------------------------------------------
測試:
>>> import bayes >>> from numpy import * >>> listOPosts,listClasses=bayes.loadDataSet() #讀取訓練資料集及類別標籤 >>> listOPosts [['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']] >>> listClasses [0, 1, 0, 1, 0, 1] >>> myVocalList=bayes.createVocabList(listOPosts) #建立詞彙表 >>> myVocalList ['cute', 'love', 'help', 'garbage', 'quit', 'I', 'problems', 'is', 'park', 'stop', 'flea', 'dalmation', 'licks', 'food', 'not', 'him', 'buying', 'posting', 'has', 'worthless', 'ate', 'to', 'maybe', 'please', 'dog', 'how', 'stupid', 'so', 'take', 'mr', 'steak', 'my'] >>> trainMat=[] >>> for postindoc in listOPosts: #將訓練資料集轉換為詞向量形式 ... trainMat.append(bayes.setOfWords2Vec(myVocalList,postindoc)) ... >>> trainMat [[0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1], [0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0], [1, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1], [0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 1, 1], [0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0]] >>> p0v,p1v,pab=bayes.trainNB0(trainMat,listClasses)#計算條件概率及侮辱性文件出現的概率p(1) >>> p0v,p1v,pab (array([-2.56494936, -2.56494936, -2.56494936, -3.25809654, -3.25809654, -2.56494936, -2.56494936, -2.56494936, -3.25809654, -2.56494936, -2.56494936, -2.56494936, -2.56494936, -3.25809654, -3.25809654, -2.15948425, -3.25809654, -3.25809654, -2.56494936, -3.25809654, -2.56494936, -2.56494936, -3.25809654, -2.56494936, -2.56494936, -2.56494936, -3.25809654, -2.56494936, -3.25809654, -2.56494936, -2.56494936, -1.87180218]), array([-3.04452244, -3.04452244, -3.04452244, -2.35137526, -2.35137526, -3.04452244, -3.04452244, -3.04452244, -2.35137526, -2.35137526, -3.04452244, -3.04452244, -3.04452244, -2.35137526, -2.35137526, -2.35137526, -2.35137526, -2.35137526, -3.04452244, -1.94591015, -3.04452244, -2.35137526, -2.35137526, -3.04452244, -1.94591015, -3.04452244, -1.65822808, -3.04452244, -2.35137526, -3.04452244, -3.04452244, -3.04452244]), 0.5) >>> >>> testEntry = ['love', 'my', 'dalmation'] >>> thisDoc = array(bayes.setOfWords2Vec(myVocalList, testEntry)) #測試資料轉詞向量 >>> thisDoc array([0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1]) >>> print testEntry,'classified as: ',bayes.classifyNB(thisDoc,p0v,p1v,pab)#判斷測試資料類別 ['love', 'my', 'dalmation'] classified as: 0
*************************************************************
二、示例:朴樹貝葉斯過濾垃圾郵件
email資料夾下有兩個資料夾ham和spam,分別存放垃圾郵件和非垃圾郵件,將這50條資料,40條作為訓練樣本,10條資料作為測試樣本,使用詞袋模型完成訓練和測試,並列印錯誤詞,計算錯誤率
>>> import bayes >>> bayes.spamTest() classification error ['benoit', 'mandelbrot', '1924', '2010', 'benoit', 'mandelbrot', '1924', '2010', 'wilmott', 'team', 'benoit', 'mandelbrot', 'the', 'mathematician', 'the', 'father', 'fractal', 'mathematics', 'and', 'advocate', 'more', 'sophisticated', 'modelling', 'quantitative', 'finance', 'died', '14th', 'october', '2010', 'aged', 'wilmott', 'magazine', 'has', 'often', 'featured', 'mandelbrot', 'his', 'ideas', 'and', 'the', 'work', 'others', 'inspired', 'his', 'fundamental', 'insights', 'you', 'must', 'logged', 'view', 'these', 'articles', 'from', 'past', 'issues', 'wilmott', 'magazine'] classification error ['yeah', 'ready', 'may', 'not', 'here', 'because', 'jar', 'jar', 'has', 'plane', 'tickets', 'germany', 'for'] classification error ['home', 'based', 'business', 'opportunity', 'knocking', 'your', 'door', 'don', 'rude', 'and', 'let', 'this', 'chance', 'you', 'can', 'earn', 'great', 'income', 'and', 'find', 'your', 'financial', 'life', 'transformed', 'learn', 'more', 'here', 'your', 'success', 'work', 'from', 'home', 'finder', 'experts'] the error rate is: 0.3 >>>
另一個例子:使用樸素貝葉斯來發現低於相關的用詞
原理:利用訓練樣本求得p0Vect——[2],p1Vect——[1],pAbusive;即P(w|ci)和P(ci),通過調整移除的高頻詞資料,達到較高的準確率,然後返回出現概率較大的詞彙。
(程式碼略)
詳細見書P70