代码之家 › 专栏 › 技术社区 › IndigoChild

Python VIF为伪变量返回无穷大的值

multicollinearity statsmodels classification numpy pandas

IndigoChild · 技术社区 · 3 年前

因此,在中风预测数据集中,我为所有分类变量创建了虚拟变量,即gender_more和gender_moemale、smoking_statussmoks和smoking_tatus_nunknown等。现在,为了检查所有变量(数字和虚拟)的多重共线性,我应用了方差膨胀函数:

import numpy as np
from statsmodels.stats.outliers_influence import variance_inflation_factor
vif_data = pd.DataFrame()

vif_data["feature"] = new_df.loc[:, new_df.columns != 'stroke'].columns
vif_data["VIF"] = [variance_inflation_factor(new_df.loc[:, new_df.columns != 'stroke'].values, i) for i in range(len(new_df.loc[:, new_df.columns != 'stroke'].columns))]
vif_data

我得到的输出如下:

feature VIF
0   age 2.836394
1   hypertension    1.111484
2   heart_disease   1.113943
3   avg_glucose_level   1.107552
4   bmi 1.342729
5   gender_Female   inf
6   gender_Male inf
7   ever_married_No inf
8   ever_married_Yes    inf
9   work_type_Govt_job  inf
10  work_type_Never_worked  inf
11  work_type_Private   inf
12  work_type_Self-employed inf
13  work_type_children  inf
14  Residence_type_Rural    inf
15  Residence_type_Urban    inf
16  smoking_status_formerly smoked  inf
17  smoking_status_never smoked inf
18  smoking_status_smokes   inf

有人能解释一下为什么伪变量的vif是无穷大吗?有没有更好的方法来检查多重共线性?谢谢