So, I've been learning about decision trees and weakest link pruning for regression from ISL and ESL. But there is a couple of things that are still unclear:
- We use the RSS (residual sum of squares) + a regularization term for our cost-function (see page 309, An Introduction to Statistical Learning for example). Is the motivation behind this that the complexity for the tree grows linearly with number of leaves/terminal nodes the trees has, or is it something deeper than that?
- What happens in cost-complexity pruning if we have two nodes to collapse/prune on the same value of alpha where one is in the branch of the other?