朴素贝叶斯算法（python 实现）

利用朴素贝叶斯贝叶斯算法实现分类

本文主要通过python构建朴素贝叶斯分类算法：

原理：

贝叶斯公式，

朴素的理解：假设个特征之间相互独立，则上面公式中:

通过此公式，可将未知的条件概率转换为已知条件概率的计算

实现步骤：

1，创建Beyes类

2，类中包括四个方法，初始化方法用来创建保存中间计算结果的容器

3，fit 方法用来计算数据长度、计算条件概率所需数据集、

P(X)=∑kP(X|Y=Yk)P(Yk)

计算P(Yk)：p_train_target

4，p_test_data用来计算一个样本的分类结果

5，classifier用来计算测试集所有分类结果

6，score用来计算准确率

import pandas as pdfrom sklearn.datasets import load_irisfrom sklearn.model_selection import train_test_splitfrom collections import Counterimport timeclass Beyes(object):def __init__(self):self.length = -1 # 保存测试集数据量self.train_target_list = [] # 目标值类别集合self.p_train_target = {} # 保存个目标值概率self.split_data_lis = [] # 保存各条件概率对应的数据集self.feature_p_lis = [] # 保存特征概率self.predict = [] # 保存分类结果def fit(self, train_data, train_target):"""P(X)=∑kP(X|Y=Yk)P(Yk)计算P(Yk)：p_train_target准备计算条件概率所需数据集：self.split_data_lis:param train_data:<arr_like>:param train_target:<arr_like>:return:"""train_length = train_data.shape[0]self.length = train_lengthtarget_list = list(set(train_target)) # 队训练集目标值去重self.train_target_list = target_list # 写入对象特征target_classifier = dict(Counter(train_target)) # 保存目标值的分类计数（字典格式）train_data = pd.DataFrame(train_data)train_data['target'] = train_target # 将数据转换为DataFrame格式方便后续聚合for target in self.train_target_list:self.p_train_target[target] = target_classifier[target]/self.length # 保存各目标值的概率split_data = train_data[train_data['target'] == target]self.split_data_lis.append(split_data)print('model had trained please use classifier() to get result')def p_test_data(self, sample):"""映射函数：取测试集的一个样本，计算各特征对应条件概率及最终的分类结果并将结果映射到测试集的target列:param sample:Serise:return self.train_target_list[position]:概率最大的类别"""result_p = []for j in range(len(self.train_target_list)):p_label = 1this_target = self.train_target_list[j]this_data = self.split_data_lis[j]for i in range(0, sample.shape[0]):feature_num_dict = dict(Counter(this_data[i])) # 计算一列数据中各类别的数量if sample[i] in feature_num_dict:label_num = feature_num_dict.get(sample[i])p_label = p_label*(label_num/this_data.shape[0]) # 计算单个特征的条件概率else:# 加入拉普拉斯平滑系数解决概率为0的情况'p_label = p_label*(1/(this_data.shape[0]+len(feature_num_dict)))this_target_p = p_label*self.p_train_target.get(this_target) # 计算该样本属于该特征的概率result_p.append(this_target_p)position = result_p.index(max(result_p)) # 概率最大的分类return self.train_target_list[position]def classifier(self, test_data):"""计算分类结果:param test_data:Serise:return:"""if self.length == -1:raise ValueError('please use fit() to train the train data set ')else:test_data = pd.DataFrame(test_data)test_data['target'] = test_data.apply(self.p_test_data, axis=1) #self.predict = list(test_data['target'])print('classfier result:', self.predict)def score(self, test_target):"""计算准确率:param test_target::return:"""if len(self.predict) == 0:raise ValueError('please use classifier() to get classifier target')else:count = 0for i in range(0, test_target.shape[0]):if test_target[i] == self.predict[i]:count += 1score = count/(test_target.shape[0])print('the Classification accuracy is:', score)if __name__ == '__main__':""";用sk_learn中的鸢尾花数据集测试"""iris = load_iris()x = iris.datay = iris.targetx_train, x_test, y_train, y_test = train_test_split(x, y)print('训练集数据量', x_train.shape)print('测试集数据量', x_test.shape)start_time = time.time()classifier = Beyes()classifier.fit(x_train, y_train)classifier.classifier(x_test)classifier.score(y_test)end_time = time.time()time_d = end_time-start_timeprint("spend time:", time_d)

程序运行结果如下