从数据帧/列表/矩阵中的循环存储系数/变量重要性

Question

I will be running a bunch of different models on a dataset in a loop and I'm seeking for a way to place the variable importance/coeffecients into a data frame for reference afterwards.

I envision a dataframe/matrix with the model name as column headers and the list of all potential variables as the row names (or vice versa).

 library(MASS)
 library(caret)

 #which to use?
 coef_df = data.frame()
 coef_list = list()

 for (i in 0:1){
 subset = Boston[which(Boston$chas==i),]
 ctrl =trainControl(method='cv',number=5)
 rf_model = train(medv ~. , data=subset, trControl=ctrl, method='rf')
 gbm_model = train(medv ~. , data=subset, trControl=ctrl, method='gbm')
 #where does this go   =varImp(rf_model)
 #where does this go   =varImp(gbm_model)
  }

I think that is more or less 90% of any coding/typing I would need to do, I just don't know who to place the variable importance values into the correct bucket in a data frame/matrix, since each time in the varImp call the variables will be in a different order - even if they might be the same here.

Thanks!

Answer 1

Central rule in R : forget the for - it is forbidden.

Now, how you can do this elegantly with data.table giving only the results for the gbm method:

gbm.DT <- Boston.DT[
  , {
    gbm_model <- train(medv ~. , data=.SD, trControl=ctrl, method='gbm')
    varImp(gbm_model)$importance
  }
  , keyby = (chas1 = chas)
  ]