
184 DIMENSIONS 1, 2, 3, . . . , ∞
training data set only. “Internal structure”here means that we are just investigating the relation
of the regressor candidates among one another, leaving the response out of discussion for now.
For a first graphical representation, we use parallel coordinates. Since the display is strongly
influenced by the sequence and the layout of the display space, we use a standard trick and show
the variables of interest twice. We make use of the alpha channel again to resolve overplotting.
Example 4.4: Parallel Plots
Input
print(parallel(fat[ -c(39, 41, 216), c(19:29,19:29)],
horizontal.axis = FALSE, scales = list(x = list(rot