Two-question tree explorer

IS 507 Lecture 10 · diabetes data, BMI and blood pressure (bp)

Which question comes first?

Show the 111 test patients

Question 1 Question 2, if yes Question 2, if no

Drag a line in the plane, move a slider, or click a curve.

Training RMSE: Test RMSE:

Tree	Question 1	Question 2, yes	Question 2, no	Training RMSE	Test RMSE

Each patient lands in one of four rectangles, and the rectangle predicts the average progression of its training patients. RMSE is the root mean squared miss. Right, top: the RMSE for every cut of question 1; the thin gray curve is question 1 alone (two rectangles), the curve a greedy tree minimizes, and the navy curve is the whole tree when question 2 is re-chosen as well as possible for that cut. Right, middle and bottom: the RMSE of the whole tree for every cut of question 2 on each side, with the other two cuts held fixed. With test patients shown, the dashed curves give the test RMSE.

Data: Efron, Hastie, Johnstone, and Tibshirani (2004), "Least Angle Regression," Annals of Statistics 32(2); the unscaled copy in scikit-learn, load_diabetes(scaled=False). The Lecture 9 split, train_test_split(test_size=0.25, random_state=507): 331 training and 111 test patients.