Naive Bayes Classifier

from collections import Counter, defaultdict
import glob, re, random, math

def tokenize(message):
    message = message.lower()
    allWords = re.findall("[a-z0-9']+", message)
    return set(allWords)

# 학습 데이:q터에서 빈도수를 세는 함수를 만들자.
# 모양은 이렇다. [스팸 메시지에 나온 빈도수, 스팸이 아닌 메시지에 나온 빈도수]
def countWords(trainingSet):
    """데이터는 메시지 내용, 스팸 여부 꼴로 구성되어 있음"""
    counts = defaultdict(lambda: [0, 0])
    for message, isSpam in trainingSet:
        for word in tokenize(message):
            counts[word][0 if isSpam else 1] += 1
    return counts

# smoothing을 통해 단어의 빈도수로 확률 값을 추정하자. > k는 가짜빈도.
# 스팸에서 단어가 나올 확률, 스팸이 아닌 메시지에서 단어가 나올 확률의 list를 반환한다.
def wordProbabilities(counts, totalSpams, totalNonSpams, k=0.5):
    """빈도수를 [단어, p(w|스팸), p(w|~스팸)] 형태로 변환"""
    return [(w,
            (spam+k)/(totalSpams+2*k),             # spam+0.5 / totalSpams+1
            (nonSpam+k)/(totalNonSpams+2*k))       # nonSpam+0.5 / totalNonSpams+1
            for w, (spam, nonSpam) in counts.items()]

# 단어가 나올 경우 / 나오지 않을 경우에 대응해 log 덧셈한 값을 return한다.
def spamProbabilities(wordProbs, message):
    messageWords = tokenize(message)
    logProbIfSpam = logProbIfNotSpam = 0.0

    # 모든 단어 반복
    for word, probIfSpam, probIfNotSpam in wordProbs:

        if word in messageWords:
            logProbIfSpam += math.log(probIfSpam)
            logProbIfNotSpam += math.log(probIfNotSpam)

        else:
            logProbIfSpam += math.log(1.0 - probIfSpam)
            logProbIfNotSpam += math.log(1.0 - probIfNotSpam)

    probIfSpam = math.exp(logProbIfSpam)
    probIfNotSpam = math.exp(logProbIfNotSpam)
    return probIfSpam / (probIfSpam+probIfNotSpam)

class naiveBayesClassifier:

    def __init__(self, k=0.5):
        self.k = k
        self.wordProbs = []

    def train(self, trainingSet):
        numSpams = len([isSpam for message, isSpam in trainingSet if isSpam])
        numNonSpams = len(trainingSet) - numSpams

        wordCounts = countWords(trainingSet)
        self.wordProbs = wordProbabilities(wordCounts, numSpams, numNonSpams, self.k)

    def classify(self, message):
        return spamProbabilities(self.wordProbs, message)

path = r"C:\path\to\spam\*\*"

def pSpamGivenWord(wordProb):
    word, probIfSpam, probIfNotSpam = wordProb
    return probIfSpam / (probIfSpam + probIfNotSpam)

def getSubjectData(path):
    data = []
    subjectRegex = re.compile(r"Subject:\s+")
    for fn in glob.glob(path):
        isSpam = "ham" not in fn

        with open(fn,'r',encoding='ISO-8859-1') as file:
                for line in file:
                    if line.startswith("Subject:"):
                        subject = subjectRegex.sub("", line).strip()
                        data.append((subject, isSpam))       
    return data              

def splitData(data, prob):
    """split data into fractions [prob, 1 - prob]"""
    results = [], []
    for row in data:
        results[0 if random.random() < prob else 1].append(row)
    return results

def trainAndTest(path):
    data = getSubjectData(path)
    random.seed(0)
    trainData, testData = splitData(data, 0.75)

    classifier = naiveBayesClassifier()
    classifier.train(trainData)

    classified = [(subject, isSpam, classifier.classify(subject))
                   for subject, isSpam in testData]

    counts = Counter((isSpam, spamProbability > 0.5)
                      for _, isSpam, spamProbability in classified)

    print(counts)
    classified.sort(key=lambda row: row[2])
    spammiestHams = list(filter(lambda row: not row[1], classified))[-5:]
    hammiestSpams = list(filter(lambda row: row[1], classified))[:5]

    print("\n")
    print("Spammiest Hams :",spammiestHams)
    print("\n")
    print("Hammiest Spams :",hammiestSpams)

    words = sorted(classifier.wordProbs, key=pSpamGivenWord)
    spammiestWords = words[-5:]
    hammiestWords = words[:5]

    print("\nSpammiest Words :",spammiestWords)
    print("\nHammiest Words :",hammiestWords)

trainAndTest(path) # 가음성 704, 진양성 101, 가양성 38, 진음성 33
>>>
Counter({(False, False): 704, (True, True): 101, (True, False): 38, (False, True): 33})


Spammiest Hams : [('Attn programmers: support offered [FLOSS-Sarai Initiative]', False, 0.9756020492343607), ('2000+ year old Greek computer reinterpreted', False, 0.9835280743958158), ('What to look for in your next smart phone (Tech Update)', False, 0.9898673229124078), ('[ILUG-Social] Re: Important - reenactor insurance needed', False, 0.9995346925825354), ('[ILUG-Social] Re: Important - reenactor insurance needed', False, 0.9995346925825354)]


Hammiest Spams : [('Re: girls', True, 0.0009520823968456288), ('Introducing Chase Platinum for Students with a 0% Introductory APR', True, 0.001256093787333424), ('.Message report from your contact page....//ytu855 rkq', True, 0.0015102442616159145), ('Testing a system, please delete', True, 0.0026908231662706177), ('Never pay for the goodz again (8SimUgQ)', True, 0.005908929345307938)]

Spammiest Words : [('year', 0.028767123287671233, 0.00022893772893772894), ('sale', 0.031506849315068496, 0.00022893772893772894), ('rates', 0.031506849315068496, 0.00022893772893772894), ('systemworks', 0.036986301369863014, 0.00022893772893772894), ('money', 0.03972602739726028, 0.00022893772893772894)]

Hammiest Words : [('spambayes', 0.0013698630136986301, 0.04601648351648352), ('users', 0.0013698630136986301, 0.036401098901098904), ('razor', 0.0013698630136986301, 0.030906593406593408), ('zzzzteana', 0.0013698630136986301, 0.029075091575091576), ('sadev', 0.0013698630136986301, 0.026785714285714284)]

# 정밀도
print(101/(101+33))
# 재현율
print(101/(101+38))
>>
0.753731343283582
0.7266187050359713