数据挖掘 —— 有监督学习（回归）

1. 线性回归模型

1.1 线性回归模型

from sklearn.linear_model import LinearRegression
LR_model = LinearRegression()

• LR_model.intercept_ ：截距
• LR_model.coef_ ：权重

1.2 岭回归模型

from sklearn.linear_model import Ridge
Ridge_model = Ridge(alpha，max_iter,tol，solver)

• alpha:正则化强度系数
• max_iter:最大迭代次数 默认1000
• tol:精度，默认1e-3
• solver：求解算法 {‘auto’, ‘svd’, ‘cholesky’, ‘lsqr’, ‘sparse_cg’, ‘sag’, ‘saga’}默认为“auto”
• random_state：int, RandomState instance, default=None 使用随机平均梯度法时使用的随机样本数
• Ridge_model.n_iter 模型性质

1.3 Lasso回归模型

from sklearn.Linear_model import Lasso
Lasso_model = Lasso(alpha，max_iter,tol，solver)

2 逻辑回归模型

from sklearn.linear_model import LogisticRegression
logistic_model = LogisticRegression(penalty,tol,C,max_iter,solver)

• penalty 为正则化的惩罚方法："l1"为一范数惩罚，"l2"为二范数惩罚
• C：默认为1，正则化强度，值越小，正则化强度越大
• solver：{‘newton-cg’, ‘lbfgs’, ‘liblinear’, ‘sag’, ‘saga’}, default=’lbfgs’

3 回归树与提升树

from sklearn.ensemble import GradientBoostingClassifier
from sklearn.ensemble import GradientBoostingRegressor
GradientBoostingClassifier(max_depth.n_estimators,learning_rate,criterion)
GradientBoostingRegressor()

• max_depth:默认为3 决定了每个决策树的的节点深度，
• learning_rate:学习速率
• criterion:优化算法

4 总结

import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression,Ridge,Lasso,LogisticRegression
from sklearn.metrics import mean_squared_error,accuracy_score,f1_score,precision_score,recall_score
from sklearn.ensemble import GradientBoostingRegressor

features = pd.read_excel("./data.xlsx",sheet_name = "features")
label = pd.read_excel("./data.xlsx",sheet_name = "label")

# ————————添加线性回归模型
# 训练集拆分
def data_split(x,y):
    X_tt,X_validation,Y_tt,Y_validation = train_test_split(x,y,test_size = 0.2)
    X_train,X_test,Y_train,Y_test = train_test_split(X_tt,Y_tt,test_size = 0.25)
    return X_train,X_validation,X_test,Y_train,Y_validation,Y_test
# 自定义回归模型评价参数
def regression_metrics(model,X_train,X_validation,X_test,Y_train,Y_validation,Y_test):
    print("train:")
    print("\tMSE:",mean_squared_error(model.predict(X_train),Y_train))
    print("validation:")
    print("\tMSE:",mean_squared_error(model.predict(X_validation),Y_validation))
    print("test:")
    print("\tMSE:",mean_squared_error(model.predict(X_test),Y_test))
# 自定义分类器判别参数
def classifier_metrics(model,X_train,X_validation,X_test,Y_train,Y_validation,Y_test):
    def self_metrics(y1,y2,name):
        print(name)
        print("\taccuracy score:",accuracy_score(y1,y2))
        print("\tf1 score:",f1_score(y1,y2))
        print("\trecall score:",recall_score(y1,y2))
        print("\tprecision score:",precision_score(y1,y2))
    self_metrics(Y_train,model.predict(X_train),"train:")
    self_metrics(Y_validation,model.predict(X_validation),"validation:")
    self_metrics(Y_test,model.predict(X_test),"test:")
    print(model.predict(X_validation))
# ————构建线性回归模型
X = features.iloc[:,[2,3]].values
Y = features.iloc[:,1].values
X_train,X_validation,X_test,Y_train,Y_validation,Y_test = data_split(X,Y)

LR_model = LinearRegression()
LR_model.fit(X_train,Y_train)
print("*"*20,"LinearRegression","*"*20)
print("coef:",LR_model.coef_)
print("intercept:",LR_model.intercept_)
regression_metrics(LR_model,X_train,X_validation,X_test,Y_train,Y_validation,Y_test)

# ————构建岭回归模型
X = features.iloc[:,[2,3]].values
Y = features.iloc[:,1].values
X_train,X_validation,X_test,Y_train,Y_validation,Y_test = data_split(X,Y)

Ridge_model = Ridge(alpha = 10,max_iter = 1e6,tol=1e-6)
Ridge_model.fit(X_train,Y_train)
print("*"*20,"RidgeRegression","*"*20)
print("coef:",Ridge_model.coef_)
print("intercept:",Ridge_model.intercept_)
regression_metrics(Ridge_model,X_train,X_validation,X_test,Y_train,Y_validation,Y_test)

# ————Lasso回归模型
X = features.iloc[:,[2,3]].values
Y = features.iloc[:,1].values
X_train,X_validation,X_test,Y_train,Y_validation,Y_test = data_split(X,Y)

lasso_model = Lasso(alpha = 0,max_iter = 1e6,tol=1e-6)
lasso_model.fit(X_train,Y_train)
print("*"*20,"LassoRegression","*"*20)
print("coef:",lasso_model.coef_)
print("intercept:",lasso_model.intercept_)
regression_metrics(lasso_model,X_train,X_validation,X_test,Y_train,Y_validation,Y_test)

# ————逻辑回归模型
X = features.values
Y = label.values
X_train,X_validation,X_test,Y_train,Y_validation,Y_test = data_split(X,Y)

logistic_model = LogisticRegression(penalty = "l2",C = 0.51)
logistic_model.fit(X_train,Y_train)
print("*"*20,"LogisticRegression","*"*20)
print("coef:",logistic_model.coef_)
print("intercept:",logistic_model.intercept_)
classifier_metrics(logistic_model,X_train,X_validation,X_test,Y_train,Y_validation,Y_test)

# ————构建梯度提升回归树模型
X = features.iloc[:,[2,3]].values
Y = features.iloc[:,1].values
X_train,X_validation,X_test,Y_train,Y_validation,Y_test = data_split(X,Y)

GBDT = GradientBoostingRegressor(n_estimators = 1000,learning_rate = 0.5)
GBDT.fit(X_train,Y_train)
print("train score:",GBDT.train_score_)
regression_metrics(GBDT,X_train,X_validation,X_test,Y_train,Y_validation,Y_test)

by CyrusMay 2022 04 05