逻辑回归--简单学习使用

logistic回归又称logistic回归分析，是一种广义的线性回归分析模型，常用于数据挖掘，疾病自动诊断，经济预测等领域。 逻辑回归根据给定的自变量数据集来估计事件的发生概率，由于结果是一个概率，因此因变量的范围在 0 和 1 之间。

以上摘选自：
https://baike.baidu.com/item/logistic%E5%9B%9E%E5%BD%92/2981575

下面进行一些简单的学习：
本次使用的数据集是乳腺癌数据集。

import pandas as pd
import numpy as np
from sklearn.datasets import load_breast_cancer
import matplotlib.pyplot as plt 
%matplotlib inline

#下载数据集
data = load_breast_cancer()
X = data.data
y = data.target

#切分训练集和测试集
from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test = train_test_split(X,y,test_size=0.3,random_state=200)

#数据标准化处理
from sklearn.preprocessing import StandardScaler
standar_X_train = StandardScaler().fit(X_train).fit_transform(X_train)
standar_X_test = StandardScaler().fit(X_train).fit_transform(X_test)

#使用网格搜索寻求最优参数
params = {"C":[*range(1,5)],
         "solver":["liblinear","newton-cg","sag","lbfgs"]}
gs = GridSearchCV(LogisticRegression(penalty="l2",max_iter=10000),param_grid=params,cv = 5)
gs.fit(standar_X_train,y_train)
gs.best_params_

#模型拟合
from sklearn.linear_model import LogisticRegression 
lr = LogisticRegression(penalty="l2",C = gs.best_params_["C"],solver=gs.best_params_["solver"],max_iter=10000).fit(standar_X_train,y_train)
[lr.score(standar_X_train,y_train),lr.score(standar_X_test,y_test)]

得到的结果是：
{‘C’: 1, ‘solver’: ‘liblinear’}
score:[0.9899497487437185, 0.9649122807017544]****

其实上述网格搜索对于C值，不够精细，不能设置成小数。为此，可以描绘学习曲线，观测不同C值下的score值的变化

#l1和l2范数训练集和测试集的分数列表
lrL1TrainList = []
lrL1TestList = []

lrL2TrainList = []
lrL2TestList = []
c_range = np.linspace(0.1,2,20)

#L1范数下只能用liblinear 坐标下降法
for i in c_range:
    lr_l1= LogisticRegression(penalty="l1",C = i,solver="liblinear",max_iter=1000).fit(Xtrian,ytrain)
    lr_l2= LogisticRegression(penalty="l2",C = i,solver="newton-cg",max_iter=1000).fit(Xtrian,ytrain)
    score_trian = lr_l2.score(Xtrian,ytrain)
    score_test  = lr_l2.score(Xtest,ytest)
    
    score_trian_1 = lr_l1.score(Xtrian,ytrain)
    score_test_1  = lr_l1.score(Xtest,ytest)
    
    lrL2TrainList.append(score_trian)
    lrL2TestList.append(score_test)
    
    lrL1TrainList.append(score_trian_1)
    lrL1TestList.append(score_test_1)
plt.plot(c_range,lrL2TrainList,color = "green",label = "l2_trian")
plt.plot(c_range,lrL2TestList,color = "blue",label = "l2_test")

plt.plot(c_range,lrL1TrainList,color = "lightgreen",label = "l1_trian")
plt.plot(c_range,lrL1TestList,color = "lightblue",label = "l1_test")

plt.legend()

绘图如下:

可以看到l2范数下的score,训练集和测试集上的分差较小，C值为1.0确实是一个不错的选择。