Fitting a Decision Tree (CART)

At each node:

1. list all possible rules (for a continuous feature, one per midpoint of consecutive sorted values),
2. score every rule and pick the best,
3. split the data through it, and
4. repeat on both halves until a stop condition is met.