Gradient Boosting: Watching a Curve Take Shape

Gradient boosting is the algorithm behind XGBoost, LightGBM and scikit-learn's GradientBoostingRegressor, and this simulator shows exactly what it does under the hood: a scatter of noisy data points sits in 3D space next to the true generating curve, and each boosting round fits one shallow regression tree — a real greedy, SSE-minimizing tree, not a canned animation — to whatever error the ensemble so far still leaves on the table. Add trees one at a time or let it auto-boost, and watch the orange prediction curve bend closer to the data while translucent walls mark each new tree's split thresholds and red residual bars shrink round by round. Shrinkage, tree depth and dataset noise are all live controls, so you can watch the same bias/variance trade-off that makes gradient boosting both powerful and easy to overfit.