Visualizes a decision tree node evaluating multiple candidate splits by computing node impurity I(node) (Gini or Entropy), selecting the split with maximum impurity reduction ΔI, then routing sample “packets” down the chosen branch to leaf nodes that predict the majority class.
3.6s animation cycle: (1) scan/evaluate split candidates, (2) highlight argmax ΔI, (3) animate sample packets flowing to children proportional to wL, (4) emphasize leaf prediction. Responsive scaling via scale=Math.min(w,h)/240, grid-snapped drawing for a retro blocky look, no external deps; impurity toggles between Gini and Entropy every two cycles.