str function mentions stats but how do I create the stat output using caret?

General
1

I have the following code:

model <- as.formula("y~z+x2+x4+x5+x6+x10+x11")
set.seed(12345)
suppressMessages(library(caret))
train.control <- trainControl(method="repeatedcv", number=10, repeats=3)
cv <- train(model,data=df2, method="glm", trControl=train.control)

cat("RMSE =",cv$results$RMSE,"\n")
cat("Rsquared =",cv$results$Rsquared,"\n")
cat("MAE =",cv$results$MAE,"\n")

The following statements result in NULL (each one)

cv$varImp
cv$aic
cv$deviance

When I see something listed in str(cv) am I to assume it can be available?
varImp, aic and deviance are included in the list.
The package documentation was sparse on this topic.

2

Hi, you can extract the AIC and deviance like this. I have used the iris dataset to make it reproducible.

model <- as.formula("Sepal.Length ~ Sepal.Width + Petal.Length")
set.seed(12345)
suppressMessages(library(caret))
train.control <- trainControl(method="repeatedcv", number=10, repeats=3)
cv <- train(model, data=iris, method="glm", trControl=train.control)

# view summary
summary(cv)

# extract these
cv$finalModel$aic
cv$finalModel$deviance

image

3

William,

Do you know why cv$ModelInfo$varImp does not work? Thanks for aic and deviance example. I tried to expand on that.

Mary

str(cv)
List of 24
 $ method      : chr "glm"
 $ modelInfo   :List of 15
  ..$ label     : chr "Generalized Linear Model"
  ..$ library   : NULL
  ..$ loop      : NULL
  ..$ type      : chr [1:2] "Regression" "Classification"
  ..$ parameters:'data.frame':	1 obs. of  3 variables:
  .. ..$ parameter: chr "parameter"
  .. ..$ class    : chr "character"
  .. ..$ label    : chr "parameter"
  ..$ grid      :function (x, y, len = NULL, search = "grid")  
  ..$ fit       :function (x, y, wts, param, lev, last, classProbs, ...)  
  ..$ predict   :function (modelFit, newdata, submodels = NULL)  
  ..$ prob      :function (modelFit, newdata, submodels = NULL)  
  ..$ varImp    :function (object, ...)  

cv$ModelInfo$varImp

Output
NULL

4

I think it is because varImp()is a function.

You can run varImp(cv) though.

image

Though maybe it should be this: cv$modelInfo$varImp(cv)

image

5

William,

Note the definition of variable importance.
variable importance - mean decrease in node impurity (and not the mean decrease in accuracy)

To me this would imply that cv$modelInfo$varImp(cv) is the correct syntax. Getting the right stats out of a str function is tricky.

This is what I tried. Note the varImp values differ.

cv$modelInfo$varImp(cv)  # 8-> 39.7
varImp(cv)               # 100 -> 0

I also looked at:

library(caret)
library(randomForest)
varImpPlot(cv,type=2) # Note: This function only works for objects of class `randomForest'

From this I would conclude that cv$modelInfo$varImp(cv) is correct. Thank you replying.

1 Like