Automatic clustering: searching for latent structures

k-means on rows and columns of the sandbox — zero overlap, two criterion families

Mind map of chapter 26: unsupervised clustering. Three typical geometries (absence, convex clusters, disc/ring), k-means on the 40 rows of M (perfect L_d/L_p agreement with overlap 𝒪 = 0 and inertia ratio I_B/I_T ≈ 73%), centroid reading, column clustering into two families (A: R, T, F↓, Tr, S, V, C cleaving; B: Dr, K↓, J_p transverse) and acknowledged limits.