第四章基于概率论的朴素贝叶斯分类

第四章基于概率论的朴素贝叶斯分类

本章主要讲基于概率论的分类方法：朴素贝叶斯分类，这里本人只是简单进行了梳理，以及代码分享。

算法介绍

测试代码

import os
import numpy as np

class Bayes:
    def __init__(self):
        pass

    def CreateDataSet(self):
        dataset = [
            ["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"]
        ]
        labelclass = [0, 1, 0, 1, 0, 1] # 0 normal word , 1 not normal word

        return dataset, labelclass

    def CreateWordList(self, dataset):
        wordlist = set([])
        for data in dataset:
            wordlist = wordlist | set(data)
        return list(wordlist)

    def WordToVec(self, wordslist, inputset):
        wordvec = [0]*len(wordslist)
        for word in inputset:
            if word in wordslist:
                wordvec[wordslist.index(word)] += 1
            else:
                print(word, "is not in Vocabulary!")
        return wordvec

    def ComputeDataSetVecMatrix(self, dataset, wordslist):
        matrix = []
        for data in dataset:
            vec = self.WordToVec(wordslist, data)
            matrix.append(vec)
        return matrix

    def TrainClassfiy(self, vecmatrix, labelvec):
        numdocs = len(vecmatrix)
        numwords = len(vecmatrix[0])
        p_cls1 = sum(labelvec) / numdocs
        p_cls0 = 1 - p_cls1
        p1num = np.ones((1, numwords))
        p1denom = 2.0
        p0num = np.ones((1, numwords))
        p0denom = 2.0
        for i in range(numdocs):
            if labelvec[i] == 1:
                p1num += vecmatrix[i]
                p1denom += sum(vecmatrix[i])
            else :
                p0num += vecmatrix[i]
                p0denom += sum(vecmatrix[i])

        p1vec = np.log(p1num / p1denom)
        p0vec = np.log(p0num / p0denom)
        return p0vec[0], p1vec[0], p_cls1

    def vec2classfiy(self, inputwordvec, p1vec, p0vec, p_cls1):
        p1 = sum(inputwordvec * p1vec) + np.log(p_cls1)
        p0 = sum(inputwordvec * p0vec) + np.log(1 - p_cls1)
        print("---------------")
        print(p1)
        if p1 > p0:
            return 1
        else:
            return 0

if __name__ == '__main__':
    bayes = Bayes()
    # create dataset
    dataset, label = bayes.CreateDataSet()

    # create vocabulary list
    wordlist = bayes.CreateWordList(dataset)
    print(wordlist)

    # word to vec
    intputset = ["my", "stop"]
    wordvec = bayes.WordToVec(wordlist, intputset)
    print(wordvec)

    # compute conditional probability
    vecmatrix = bayes.ComputeDataSetVecMatrix(dataset, wordlist)
    print(vecmatrix)

    # train classfiy
    p0vec, p1vec, p_cls1 = bayes.TrainClassfiy(vecmatrix, label)
    print(p0vec, p1vec, p_cls1)

    # wordvec calssfiy
    # inputwords = ["my", "love", "my", "dog"]
    inputwords = ["stupid", "my", "dog"]
    wordvec = bayes.WordToVec(wordlist, inputwords)
    res = bayes.vec2classfiy(wordvec, p1vec, p0vec, p_cls1)
    print("res: ", res)