Hi!
I was expecting some discussion of scaling labels in regression problems, in chapters 2 or 3.
but I don't see any such discussion. Am I missing it? or does it not exist?
In many real world regression problems, the labels are very skewed (and maybe bounded below by 0, like income amounts for example).
In that case it is often better for both learning and evaluation to log the labels (and maybe normalize to [0,1] or scale to mean=0 and sd=1) before assigning them as the target in the task.
Does mlr3 have support for this? or is that left up to the user to do it yourself before Task initialization?