sklearn中回归问题的评估方法

在对回归问题的建模分析中，经常会遇到对回归问题的评估问题，如何评估回归模型的优劣呢，本文整理了sklearn中的metrics中关于回归问题的评估方法。

首先导入相应的函数库并建立模型

#导入相应的函数库
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error
from sklearn.metrics import median_absolute_error
from sklearn.metrics import mean_squared_log_error
from sklearn.metrics import mean_absolute_error
import numpy as np

#加载数据
bos_house = datasets.load_boston()
bos_house_data = bos_house['data']
bos_house_target = bos_house['target']

#建模分析
x_train,x_test,y_train,y_test = train_test_split(bos_house_data,bos_house_target,random_state=41)
forest_reg = RandomForestRegressor(random_state=41)
forest_reg.fit(x_train,y_train)
y_pred = forest_reg.predict(x_test)

1、mean_squared_error（MSE 常用）

简称MSE，即均方误差，计算公式为：

$MSE=\frac{1}{n}{\sum }_{i=1}^n(y_i-\widehat{y_i}{)}^2$

一般使用RMSE进行评估（这个回归分析模型中最常用的评估方法）：

$RMSE=\sqrt{\frac{1}{n}{\sum }_{i=1}^n(y_i-\widehat{y_i}{)}^2}$

print('MSE为：',mean_squared_error(y_test,y_pred))
print('MSE为(直接计算)：',np.mean((y_test-y_pred)**2))
print('RMSE为：',np.sqrt(mean_squared_error(y_test,y_pred)))

#结果为
#MSE为： 20.2051377953
#MSE为(直接计算)： 20.2051377953
#RMSE为： 4.49501254673

2、mean_squared_log_error

计算公式为：

$MSLE=\frac{1}{n}{\sum }_{i=1}^n{\left(log(y_i+1)-log(\widehat{y_i}+1)\right)}^2$

print(mean_squared_log_error(y_test,y_pred))
print(np.mean((np.log(y_test+1)-np.log(y_pred+1))**2))

#0.0387712039735
#0.0387712039735

3、median_absolute_error

计算公式为：

$median(|y_i-\widehat{y_i}|)$

print(median_absolute_error(y_test,y_pred))
print(np.median(np.abs(y_test-y_pred)))

#1.79
#1.79

4、mean_absolute_error(MAE)

计算公式为：

$\frac{1}{n}{\sum }_{i=1}^n|y_i-\widehat{y_i}|$

print(mean_absolute_error(y_test,y_pred))
print(np.mean(np.abs(y_test-y_pred)))

#2.7005511811
#2.7005511811

5、explained_variance_score

解释方差的得分，计算公式为：

$1-\frac{var(y-\hat{y})}{var(y)}$

print(explained_variance_score(y_test,y_pred))
print(1-np.var(y_test-y_pred)/np.var(y_test))

#0.571770465868
#0.571770465868

6、r2_score

计算公式为：

$$R^2(y,\hat{y})=1-\frac{{\sum }_{i=0}^n(y_i-\widehat{y_i}{)}^2}{{\sum }_{i=1}^n(y_i-\bar{y}{)}^2}$$
$$其中\bar{y}=\frac{1}{n}\sum _{i=1}^n{y}_i$$

print(r2_score(y_test,y_pred))
print(1-(np.sum((y_test-y_pred)**2))/np.sum((y_test -np.mean(y_test))**2))

#0.566962879097
#0.566962879097

附：所有代码如下：

from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error
from sklearn.metrics import median_absolute_error
from sklearn.metrics import mean_squared_log_error
from sklearn.metrics import mean_absolute_error
from sklearn.metrics import explained_variance_score
from sklearn.metrics import r2_score

import numpy as np


bos_house = datasets.load_boston()
bos_house_data = bos_house['data']
bos_house_target = bos_house['target']

x_train,x_test,y_train,y_test = train_test_split(bos_house_data,bos_house_target,random_state=41)
forest_reg = RandomForestRegressor(random_state=41)
forest_reg.fit(x_train,y_train)
y_pred = forest_reg.predict(x_test)

#mean_squared_error
print('MSE为：',mean_squared_error(y_test,y_pred))
print('MSE为(直接计算)：',np.mean((y_test-y_pred)**2))

print('RMSE为：',np.sqrt(mean_squared_error(y_test,y_pred)))

#median_absolute_error
print(np.median(np.abs(y_test-y_pred)))
print(median_absolute_error(y_test,y_pred))

#mean_absolute_error
print(np.mean(np.abs(y_test-y_pred)))
print(mean_absolute_error(y_test,y_pred))

#mean_squared_log_error
print(mean_squared_log_error(y_test,y_pred))
print(np.mean((np.log(y_test+1)-np.log(y_pred+1))**2))

#explained_variance_score
print(explained_variance_score(y_test,y_pred))
print(1-np.var(y_test-y_pred)/np.var(y_test))

#r2_score
print(r2_score(y_test,y_pred))
print(1-(np.sum((y_test-y_pred)**2))/np.sum((y_test -np.mean(y_test))**2))

参考资料

机器学习中的一些COST函数的说明