sklearn -- 线性回归

线性回归

在统计学中，线性回归(Linear Regression)是利用称为线性回归方程的最小平方函数对一个或多个自变量和因变量之间关系进行建模的一种回归分析。这种函数是一个或多个称为回归系数的模型参数的线性组合。只有一个自变量的情况称为简单回归,大于一个自变量情况的叫做多元回归。（这反过来又应当由多个相关的因变量预测的多元线性回归区别，而不是一个单一的标量变量。）

简单线性回归

损失函数

通过对损失函数求导，得到模型最优的参数：a,b

python代码 数据处理、模型训练和预测

	import numpy as np
	from sklearn import datasets
	boston = datasets.load_boston()
	
	X = boston.data
	y = boston.target
	x = X[:,5] //取第五个属性的值
	x = x[y < 50.0]
	y = y[y< 50.0]
	x = x.reshape(-1,1) //将数组转为矩阵
	y = y.reshape(-1,1)//将数组转为矩阵
	
	from sklearn.model_selection import train_test_split
	X_train,X_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=686)
	from sklearn.linear_model import LinearRegression
	lin_reg = LinearRegression()
	lin_reg.fit(X_train,y_train)
	
	y_predict = lin_reg.predict(X_test)

多元线性回归

多元线性回归模型

损失函数

python 代码

X_mul = boston.data
y_mul = boston.target
x_mul = X_mul[y_mul < 50.0]
y_mul = y_mul[y_mul< 50.0]
from sklearn.model_selection import train_test_split
X_train,X_test, y_train, y_test = train_test_split(x_mul, y_mul, test_size=0.2, random_state=686)
from sklearn.linear_model import LinearRegression
lin_reg = LinearRegression()
lin_reg.fit(X_train,y_train)
y_predict = lin_reg.predict(X_test)
y_predict

线性回归算法的评测

MSE、RMSE、MAE、R Squared等评测指标

MSE(均方误差 Mean Squared Error)

mse_test = np.sum((y_predict - y_test)**2) / len(y_test)
// 29.10250291674462


//sklearn 中的MSE(均方误差 Mean Squared Error)
from sklearn.metrics import mean_squared_error
mean_squared_error(y_test,y_predict)
//29.10250291674462

RMSE(均方根误差 Root Mean Squared Error)

from math import sqrt
rmse_test = sqrt(mse_test)
// 5.394673569062786

MAE(平均绝对误差 Mean Absolute Error)

mae_test = np.sum(np.absolute(y_predict - y_test)) / len(y_test)
// 3.9705657868108397


// sklearn 中的MAE(平均绝对误差 Mean Absolute Error)
from sklearn.metrics import mean_absolute_error
mean_absolute_error(y_test,y_predict)
//3.9705657868108397

R Squared

1 - mean_squared_error(y_test,y_predict) / np.var(y_test)
//0.5297519058857616

// sklearn 中的R Squared
from sklearn.metrics import r2_score
r2_score(y_test,y_predict)
//0.5297519058857616