Fig. 15

Comparison of MLS and clustering weights for random (black points), k-medoids (non-normalized blue and normalized red points) and linear (green points) clusterings for \(k=15\) clusters. minimum linear  (olive point) is a manually constructed linear clustering with the smallest possible MLS (defined in the "Colorings inferred from the adjacency graph" section). optimal (orange and brown) points are clusterings with 0 MLS (defined in the "Divergence from linearity" section). The mean MLS value is 35 for random, 19.5 for linear, 20.0 for blue k-medoids points, and 18.4 for red k-medoids points