4.19【机器学习】boosting代码

H（x)是集成分类的结果，h(x)是单个基分类的结果，把他们都加起来然后经过一个sign函数，看是否超过半数

Hx≠fx就是说集成分类是错的，那么就是至少有一半的基分类器答错了，或者说是最多一半的基分类器答对了

import numpy as np
import pandas as pd
from sklearn.utils.multiclass import type_of_target
from collections import namedtuple
def train_nb(X, y):
	m, n = X.shape#m是数据集数据总数，n是特征数量
	p1 = (len(y[y == '是']) + 1) / (m + 2)  # 拉普拉斯平滑，类C为好瓜，分母为样本总数+2
	p1_list = []  # 用于保存正例下各属性的条件概率
	p0_list = []#反例下各属性的条件概率
	X1 = X[y == '是']
	X0 = X[y == '否']#得到两类样本对应X的所在行数，即对应起来
	m1, _ = X1.shape#正例数量
	m0, _ = X0.shape#反例数量
	for i in range(n):#开始遍历数据的n个属性
		xi = X.iloc[:, i]#得到每个样本的属性值，这是一列的，m*1向量
		p_xi = namedtuple(X.columns[i], ['is_continuous', 'conditional_pro'])  # 用于储存每个变量的情况
		#定义出一个元组，第一项是属性特征的名字，第二项为是否为连续值，以及其概率
		is_continuous = type_of_target(xi) == 'continuous'
		xi1 = X1.iloc[:, i]#得到正反类里在特征i上的取值
		xi0 = X0.iloc[:, i]
		if is_continuous:  # 连续值时，conditional_pro 储存的就是 [mean, var] 即均值和方差
			xi1_mean = np.mean(xi1)
			xi1_var = np.var(xi1)
			xi0_mean = np.mean(xi0)
			xi0_var = np.var(xi0)
			p1_list.append(p_xi(is_continuous, [xi1_mean, xi1_var]))
			p0_list.append(p_xi(is_continuous, [xi0_mean, xi0_var]))
		else:  # 离散值时直接计算各类别的条件概率
			unique_value = xi.unique()  # 取值情况
			nvalue = len(unique_value)  # 取值个数

			xi1_value_count = pd.Series(xi1).value_counts()[unique_value].fillna(0) + 1  # 计算正样本中，该属性每个取值的数量，并且加1，即拉普拉斯平滑
			xi0_value_count = pd.Series(xi0).value_counts()[unique_value].fillna(0) + 1

			p1_list.append(p_xi(is_continuous, np.log(xi1_value_count / (m1 + nvalue))))
			p0_list.append(p_xi(is_continuous, np.log(xi0_value_count / (m0 + nvalue))))
			return p1, p1_list, p0_list

def predict_nb(x, p1, p1_list, p0_list):
	n = len(x)
	x_p1 = np.log(p1)
	x_p0 = np.log(1 - p1)
	for i in range(n):
		p1_xi = p1_list[i]
		p0_xi = p0_list[i]
		if p1_xi.is_continuous:
			mean1, var1 = p1_xi.conditional_pro
			mean0, var0 = p0_xi.conditional_pro
			x_p1 += np.log(1 / (np.sqrt(2 * np.pi) * var1) * np.exp(- (x[i] - mean1) ** 2 / (2 * var1 ** 2)))
			x_p0 += np.log(1 / (np.sqrt(2 * np.pi) * var0) * np.exp(- (x[i] - mean0) ** 2 / (2 * var0 ** 2)))
		else:
			x_p1 += p1_xi.conditional_pro[x[i]]
			x_p0 += p0_xi.conditional_pro[x[i]]
		if x_p1 > x_p0:
			return '是'
		else:
			return '否'
if __name__ == '__main__':
	data_path = r'D:\MachineLearning\pro1\watermelon3_0_Ch.csv'
	data = pd.read_csv(data_path, index_col=0)
	X=data.iloc[:,:-1]
	y=data.iloc[:,-1]
	p1, p1_list, p0_list = train_nb(X, y)
	x_test = X.iloc[0, :]
	print(predict_nb(x_test, p1, p1_list, p0_list))