new tree density $\propto$ base distribution $\times$ past modifier (distance to cut trees and existing trees) Use SVM to distinguish good / bad trees? With some "similarity" kernel