比较样本方差对线性拟合结果的影响

代码和结果如下：

import numpy as np
import pandas as pd
from sklearn.linear_model import LinearRegression
import seaborn as sns

def self_func(i):
    
    np.random.seed(i)
    n = 101
    x = np.linspace(0,10,n)
    noise = np.random.randn(n)
    
    y1 = 2.5 * x + 0.8 + 20.0 * noise
    y2 = 2.5 * x + 0.8 + 2.0 * noise
    
    lr1 = LinearRegression()
    lr1.fit(x.reshape(-1,1),y1)
    
    lr2 = LinearRegression()
    lr2.fit(x.reshape(-1,1),y2)

    return pd.DataFrame([[0,float(lr1.coef_)],[1,float(lr2.coef_)]],columns=['cate','coef_'])

result = self_func(0)
for i in range(100):
    df = self_func(i)
    result = result.append(df,ignore_index=True)

sns.stripplot(x="cate", y="coef_", data=result,jitter=True)
    
df1 = result[result['cate'].isin([0])]
df2 = result[result['cate'].isin([1])]

print (df1.describe())
print (df2.describe())

结果：

         cate       coef_
count    101.0  101.000000
mean     0.0      2.658496
std      0.0      0.641301
min      0.0      1.183393
25%      0.0      2.146773
50%      0.0      2.584440
75%      0.0      3.147349
max      0.0      4.042992
         cate       coef_
count    101.0  101.000000
mean     1.0      2.515850
std      0.0      0.064130
min      1.0      2.368339
25%      1.0      2.464677
50%      1.0      2.508444
75%      1.0      2.564735
max      1.0      2.654299