线性回归、交叉验证、最优调参

一、普通的线性模型

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

from sklearn.linear_model import LinearRegression

from sklearn.model_selection import train_test_split

from sklearn.preprocessing import StandardScaler

from sklearn import metrics

%matplotlib inline

data = pd.read_csv('Advertising.csv',index_col=0)#第一列为index

data.head()

#切分训练集和测试集

x = data.values[:,:3]

y = data.values[:,3]

x_train,x_test,y_train,y_test = train_test_split(x,y,train_size=0.7,random_state=0)

#标准化处理

sc = StandardScaler()

x_train_std = sc.fit_transform(x_train)

x_test_std = sc.transform(x_test)

#训练模型

linreg = LinearRegression()

linreg.fit(x_train_std,y_train)

y_pred = linreg.predict(x_test_std)

#检验模型结果

mse = np.average((y_pred-y_test)**2)

metrics.mean_squared_error(y_pred,y_test)  #这个也是均方误差

r2 = metrics.r2_score(y_test,y_pred)  #R2值，注意参数，前面的是实际值，后面的是预测值

mse,r2

#计算R2

def calculater2(y_pred,y_test):

    RSS = ((y_pred-y_test)**2).sum()

    TSS = (((y_test-np.average(y_test))**2)).sum()

    return 1-(RSS/TSS)

calculater2(y_pred,y_test)

#画图

fig = plt.figure(figsize=(10,6))

plt.plot(y_test)

plt.plot(y_pred)

二、加入正则化的模型

Ridge回归

from sklearn.linear_model import RidgeCV,LassoCV  #用这个自带交叉验证参数

from sklearn.model_selection import GridSearchCV  #如果使用RidgeCV就不用GridSearchCV这个API了

#使用RidgeCV来建立参数

alpha = np.logspace(-3,2,10)    #生成超参数，10的-3次方到10的2次方的等差数列

ridge = RidgeCV(alpha,cv=5)

ridge.fit(x_train_std,y_train)

ridge.alpha_  #输出超参数的值

#使用Ridge配合GridSearchCV来做

from sklearn.linear_model import Ridge,Lasso

ridge_model = GridSearchCV(Ridge(),param_grid={'alpha':alpha},cv=5)

ridge_model.fit(x_train_std,y_train)

ridge_model.best_params_

#验证模型效果

y_pred_ridge = ridge.predict(x_test_std)

mse_ridge = metrics.mean_squared_error(y_test,y_pred_ridge)

r2_ridge = metrics.r2_score(y_test,y_pred_ridge)

mse_ridge,r2_ridge

Lasso回归

#建立模型

lasso = LassoCV(alphas=alpha,cv=5)

lasso.fit(x_train_std,y_train)

lasso.alpha_

#验证模型效果

y_pred_lasso = lasso.predict(x_test_std)

mse_lasso = metrics.mean_squared_error(y_test,y_pred_lasso)

r2_lasso = metrics.r2_score(y_test,y_pred_lasso)

mse_lasso,r2_lasso