▶ We need to know this to decide where to split How many splits should we try? ▶ We used 100 before - is there a better choice? When do we stop splitting?
The terminal nodes each represent an interval of the input variable(s). We choose to represent the outcome in this interval by a constant function. As for most regression problems, we say that a constant function is good if it has low residual sum of squares when compared to training outcome data.