w**********n 发帖数: 5 | 1 做多元回归,共有4.50个自变量,500多个obs,但自变量之间有很强的共线性。用
elastic net方法做model selection,试了试一下子模型里有很多变量。要怎样可以筛
选这些变量呢?BIC value怎么算?相关R的命令怎么写? | G*****1 发帖数: 137 | 2 a simple way is to get rid of some of them.
【在 w**********n 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 做多元回归,共有4.50个自变量,500多个obs,但自变量之间有很强的共线性。用 : elastic net方法做model selection,试了试一下子模型里有很多变量。要怎样可以筛 : 选这些变量呢?BIC value怎么算?相关R的命令怎么写?
| D******n 发帖数: 2836 | 3 ya, if there is some prior knowledge ,just choose the ones u think are
important. Maybe 20-30 of them, then you can use forward/backward selection
methods, LARS, LASSO.
Or you can use PCA to reduce the dimension of the data.
Maybe directly using LASSO or Forward/Backward selection methods can work
too, but not sure.
【在 G*****1 的大作中提到】![](/moin_static193/solenoid/img/up.png) : a simple way is to get rid of some of them.
|
|