11 Gradient Boosting

Final MSE: 291.537
h_1 thresholds: [np.float64(5.336134433746338), np.float64(7.26890754699707), np.float64(8.78151273727417)]
h_2 thresholds: [np.float64(5.92436957359314), np.float64(7.689075708389282), np.float64(9.201680660247803)]
h_3 thresholds: [np.float64(3.9075629711151123), np.float64(6.7647058963775635), np.float64(8.361344337463379)]
h_4 thresholds: [np.float64(4.495798349380493), np.float64(8.0252103805542), np.float64(9.201680660247803)]
h_5 thresholds: [np.float64(3.2352941036224365), np.float64(6.5126049518585205), np.float64(9.033613204956055)]
h_6 thresholds: [np.float64(2.815126061439514), np.float64(6.848739385604858), np.float64(8.529411792755127)]
(a) Sequential Gradient Boosting: Each adaptive weak learner (depth‑2 tree) targets new residual structure
(b)
(c)
(d)
(e)
(f)
(g)
Figure 11.1