[Bostonian]

Looking at figure 20 of the blog, my feeling is that the response surface produced the decision tree is too jagged and overfit. Maybe I overestimate the importance of smoothness in the function fit?