统计建模：Logistic回归模型（逻辑回归）

Logistic回归（逻辑回归）：用量化特征预测某事发生的概率，取值范围是0到1，多用于二分类问题。

$f=\frac{1}{1+e^{-t}}$

• 输入t：样本特征中的线性函数
• 几率（Odds）：设某事件发生的概率为 P，则该事件的几率为该事件发生的概率与不发生的概率之比
  $Odds=\frac{P}{1-P}Odds\in [0，+\infty )$
• 可基于样本的特征构建线性函数，函数值对应时间的对数几率 $lo{g}_e(\frac{P}{1-P})={\beta }_0+{\beta }_{1}x$其值域为$(-\infty ,+\infty )$

bikeshare数据集：http://archive.ics.uci.edu/ml/datasets/Bike+Sharing+Dataset

import pandas as pd
import numpy as np
from matplotlib import pyplot as plt

table = pd.DataFrame({'prob':[0.01,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,0.99]})
table['odds']=table['prob']/(1-table['prob'])
table['log-odds']=np.log(table['odds'])
plt.plot(table['prob'],'g')
plt.plot(table['odds'],'y')
plt.plot(table['log-odds'],'m')
plt.legend({'probability', 'Odds', 'log_odds'})

plt.ylim(-6, 6)

$P=\frac{e^{{\beta }_0+{\beta }_{1}x}}{1+e^{{\beta }_0+{\beta }_{1}x}}$

• Logistic回归就是对几率作线性回归
• 优化准则：极大化所有样本的对数似然函数
• 对于非数值型数据：使用 one-hot 编码

import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
from sklearn.linear_model import LogisticRegression
from sklearn import metrics
from scipy import stats
from sklearn.model_selection import train_test_split


bikes= pd.read_csv('E:\my_data\\bikeshare\hour.csv')feature_cols=['temp']

x=bikes[feature_cols]
bikes['above_average']=bikes['cnt']>=bikes['cnt'].mean()
y=bikes['cnt']>=bikes['cnt'].mean()

x_train, x_test, y_train, y_test=train_test_split(x, y)
logreg=LogisticRegression()
logreg.fit(x_train, y_train)

print(pd.DataFrame(np.transpose([y_test.values, logreg.predict(x_test)]), columns ={'真实值','预测值'}))
print('\n')
print('分类准确率是：', logreg.score(x_test, y_test))

when_dummies=pd.get_dummies(bikes['season'],prefix='season_')
when_dummies=when_dummies.iloc[:,1:]

new_bikes=pd.concat([bikes['temp'], when_dummies],axis=1)

x = new_bikes
x_train, x_test, y_train, y_test=train_test_split(x, y)
logreg=LogisticRegression()
logreg.fit(x_train, y_train)

print('用气温、季节作为预测自变量，预测的准确率是：', logreg.score(x_test, y_test))