不太可能恢复拟合或值(另请参阅this question)。您知道自己适合什么,然后绘制残差也是有道理的。下面我使用一个示例数据集:
import matplotlib. pyplot as plt
import seaborn as sns
import numpy as np
import statsmodels.api as sm
df_advertising = pd.DataFrame({'Radio':np.random.randint(1,10,100)})
df_advertising['Sales'] = 3*df_advertising['Radio'] + np.random.normal(10,4,100)
我们可以使用 seaborn 来绘制它:
fig,axes = plt.subplots(1,2,figsize=(10,5))
sns.regplot(x = 'Radio',y='Sales',data=df_advertising,ax = axes[0])
sns.residplot(x = 'Radio',y='Sales',data=df_advertising,ax = axes[1])
seaborn 使用 statsmodels,所以让我们使用它来拟合并获得预测:
mod = sm.OLS(df_advertising['Sales'],sm.add_constant(df_advertising['Radio']))
res = mod.fit()
test = df_advertising[['Radio']].drop_duplicates().sort_values('Radio')
predictions = res.get_prediction(sm.add_constant(test))
predictions = pd.concat([test,predictions.summary_frame(alpha=0.05)],axis=1)
predictions.head()
Radio mean mean_se mean_ci_lower mean_ci_upper obs_ci_lower obs_ci_upper
13 1 11.132902 0.700578 9.742628 12.523175 3.862061 18.403742
6 2 14.480520 0.582916 13.323742 15.637298 7.250693 21.710347
2 3 17.828139 0.478925 16.877728 18.778550 10.628448 25.027829
4 4 21.175757 0.399429 20.383104 21.968411 13.995189 28.356326
10 5 24.523376 0.360990 23.807002 25.239750 17.350827 31.695924
在上面,我创建了不重复数据点的测试(因为我的是计数)。现在我们可以绘制所有内容了。残差就在 statsmodels 对象的resid 之下:
fig,axes = plt.subplots(1,3,figsize=(15,5))
sns.scatterplot(x='Radio',y='Sales',ax=axes[0],data=df_advertising)
axes[0].plot(predictions['Radio'], predictions['mean'], lw=2)
axes[0].fill_between(x=predictions['Radio'],
y1=predictions['mean_ci_lower'],y2=predictions['mean_ci_upper'],
facecolor='blue', alpha=0.2)
sns.scatterplot(x='Radio',y='Sales',ax=axes[1],
data=pd.DataFrame({'Radio':df_advertising['Radio'],
'Sales':res.resid})
)
axes[1].axhline(0, ls='--',color="k")
sns.distplot(res.resid,ax=axes[2],bins=20)