NaiveBayes算法设计

标签：num probability conditional feature len NaiveBayes 算法 train 设计

一、朴素贝叶斯算法公式：

二、程序设计：

import numpy as np

#构造NB分类器

def Train(X_train, Y_train, feature):

    global class_num,label

    class_num = 2           #分类数目

    label = [1, -1]         #分类标签

    feature_len = 3         #特征长度

    #构造3×2的列表

    feature = [[1, 'S'],    

               [2, 'M'],

               [3, 'L']]

    prior_probability = np.zeros(class_num)                         # 初始化先验概率

    conditional_probability = np.zeros((class_num,feature_len,2))   # 初始化条件概率

    

    positive_count = 0     #统计正类

    negative_count = 0     #统计负类

    for i in range(len(Y_train)):

        if Y_train[i] == 1:

            positive_count += 1

        else:

            negative_count += 1

    prior_probability[0] = positive_count / len(Y_train)    #求得正类的先验概率

    prior_probability[1] = negative_count / len(Y_train)    #求得负类的先验概率

    

    '''

    conditional_probability是一个2*3*2的三维列表，第一维是类别分类，第二维和第三维是一个3*2的特征分类

    '''

    #分为两个类别

    for i in range(class_num):

        #对特征按行遍历

        for j in range(feature_len):

            #遍历数据集，并依次做判断

            for k in range(len(Y_train)):

                if Y_train[k] == label[i]: #相同类别

                    if X_train[k][0] == feature[j][0]:

                        conditional_probability[i][j][0] += 1

                    if X_train[k][1] == feature[j][1]:

                        conditional_probability[i][j][1] += 1

    class_label_num = [positive_count, negative_count]  #存放各类型的数目

    for i in range(class_num):

        for j in range(feature_len):

            conditional_probability[i][j][0] = conditional_probability[i][j][0] / class_label_num[i]  #求得i类j行第一个特征的条件概率

            conditional_probability[i][j][1] = conditional_probability[i][j][1] / class_label_num[i]  #求得i类j行第二个特征的条件概率

    return prior_probability,conditional_probability

#给定数据进行分类

def Predict(testset, prior_probability, conditional_probability, feature):

    result = np.zeros(len(label))

    for i in range(class_num):

        for j in range(len(feature)):

            if feature[j][0] == testset[0]:

                conditionalA = conditional_probability[i][j][0]

            if feature[j][1] == testset[1]:

                conditionalB = conditional_probability[i][j][1]

        result[i] = conditionalA * conditionalB * prior_probability[i]

    result = np.vstack([result,label])

    return result

def main():

    X_train = [[1, 'S'], [1, 'M'], [1, 'M'], [1, 'S'],  [1, 'S'],

               [2, 'S'], [2, 'M'], [2, 'M'], [2, 'L'],  [2, 'L'],

               [3, 'L'], [3, 'M'], [3, 'M'], [3, 'L'],  [3, 'L']]

    Y_train = [-1, -1, 1, 1, -1, -1, -1, 1, 1, 1, 1, 1, 1, 1, -1]   

    #构造3×2的列表

    feature = [[1, 'S'],    

               [2, 'M'],

               [3, 'L']]

    testset = [2, 'S']

    

    prior_probability, conditional_probability= Train(X_train, Y_train, feature)

    

    result = Predict(testset, prior_probability, conditional_probability, feature)

    print(result)

if __name__ == '__main__':

    main()

三、实验结果：

标签：num,probability,conditional,feature,len,NaiveBayes,算法,train,设计
From： https://blog.csdn.net/2302_77382298/article/details/140216696