Side-by-side visualization of two unsupervised clustering approaches. Left panel animates K-means: points are assigned to the nearest centroid (mu_j), then centroids move to the mean of their assigned points, reducing within-cluster sum of squared errors (SSE) shown as bars. Right panel animates hierarchical agglomerative clustering: a dendrogram is built by repeatedly merging the closest clusters under single-linkage, with the currently selected merge highlighted both in the dendrogram and on the point scatter plot.
Deterministic synthetic dataset and animations. K-means uses alternating assignment (nearest centroid) and centroid update (mean) with eased interpolation; SSE bars update each cycle (4.2s). Hierarchical clustering precomputes single-linkage agglomerative merges and draws them incrementally as a dendrogram; current merge is highlighted and also shown as a nearest cross-cluster point pair. Blocky aesthetic uses grid snapping (4px base) and a tiny 3x5 pixel font for labels.