Access to shape functions in mboost

Question

I am looking into the mboost package in R for creating GAM models. Below is code for creating a very simple GAM model with tree stumps on the mtcars data set.

library(mboost)

my_model = mboost(formula = mpg ~ cyl + disp + hp,
                  data = mtcars, 
                  baselearner = "btree", 
                  control = boost_control(mstop = 1000, nu = 0.1))

par(mfrow = c(1,3))
plot(my_model)

The resulting shape plots for the three functions $f_{cyl}, f_{disp}, f_{hp}$ look like this:

I can now use the model to predict new instances by calling predict(mymodel, newdata). However, is there a way to predict by using one of the shape functions at the time? In practice, this means that I would like to call something like predict(f_cyl, newdata), predict(f_disp, newdata) or predict(f_hp, newdata) where f_cyl, f_disp and f_hp are the shape functions that together make up the GAM.

Answer 1

The solution was to use predict.mboost(object = my_model, newdata = mtcars, which = 1), in order to get predictions only from shape function 1 (and which = 2 to get predictions only from shape function 2, etc.).

Now these predictions can be stored in a data frame and can plotted in ggplot2, or used for other purposes.

Thanks to Dr. Benjamin Hofner for pointing this out to me!