Label +1
Label −1
Current stump cut
drag to pan · scroll to zoom
εt (red) / αt (teal) per round
AdaBoost builds a strong classifier out of many weak ones by training them one at a time and reweighting the training set after each round: samples the current committee gets wrong get heavier, forcing the next weak learner to focus on exactly what the ensemble is still missing. This simulator renders that process on an XOR-like 2D dataset (not solvable by any single axis-aligned cut) — dot size tracks each sample's live weight, a highlighted stump line appears every round with its computed weighted error ε and learner weight α, and the background shades in the combined weighted-vote decision region as rounds accumulate, so you can watch a strong boundary emerge from a stack of individually weak ones. Pan and zoom the plane yourself, and watch the ε/α history strip converge round by round.