携程笔试的时候碰到了这个题目，当时其实没多想。贝叶斯这个路子怕也太过气了吧... 携程也真是...

回顾思路

• 计算先验概率
• 计算条件概率
• 不同类别概率估计

原始数据集

image

代码

加载数据集

import numpy as np

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']]
    classVec = [0,1,0,1,0,1]    #1 is abusive, 0 not
    return postingList,classVec

这里类别为两类，1-恶意留言；0-非恶意留言。

vocab

def getVocabList(dataSet):
    vocab = {}
    vocab_reverse = {}
    index = 0
    for line in dataSet:
        for word in line:
            if word not in vocab:
                vocab[word] = index
                vocab_reverse[index] = word
                index += 1
    return vocab,vocab_reverse

先验概率与条件概率

def native_bayes(vocab,postingList,classVec):
    # 先验概率
    label = [0,1]
    label_num = len(label)
    vocab_len = len(vocab)

    prior_probability = np.ones(label_num)                     # 初始化先验概率
    conditional_probability = np.ones((label_num,vocab_len))   # 初始化条件概率
    postingList_ids = [[vocab[word] for word in line]for line in postingList]
    # 默认N为2,
    p_n = np.array([2,2])

    for i in range(len(postingList_ids)):
        for word in postingList_ids[i]:
            conditional_probability[classVec[i]][word]+=1
            p_n[classVec[i]] += 1

    # 条件概率
    conditional_probability[0] /= p_n[0]  
    conditional_probability[1] /= p_n[1]  

    # 先验概率
    all_N = sum(p_n)
    p_n = p_n/all_N
    return p_n,conditional_probability

argmax 判断

def judge(testEntry):
    postingList,classVec = loadDataSet()
    vocab,vocab_reverse = getVocabList(postingList)
    p_n,conditional_probability = native_bayes(vocab,postingList,classVec)
    Ans_p = p_n
    
    testEntry_ids = [vocab[word] for word in testEntry]
    for num in testEntry_ids:
        Ans_p[0] *= conditional_probability[0][num]
        Ans_p[1] *= conditional_probability[1][num]
    return np.argmax(Ans_p)

调用

judge(testEntry = ['stupid', 'garbage'])

输出 1，和我们预期的一样。