Rousseeuw silhouette in functional L2 space

Computes Rousseeuw's silhouette (Rousseeuw, 1987) for a partition of keywords on the smoothed occurrence curves \(f_j(t)\). The within and between distances are pairwise functional \(L_2\) distances. Singleton classes contribute zero, following Rousseeuw's convention.

Usage

ljmds.silhouette(f, cl)

ljmds.silhouette.per.keyword(f, cl)

Arguments

f

Numeric matrix of size n x p; column j is the smoothed occurrence curve of keyword j.

cl

Integer vector of length p with cluster assignments.

Value

Mean silhouette width (scalar) or a per-keyword vector.

References

Rousseeuw, P.J. (1987) Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20, 53–65.

See also

ljmds.pipeline() which produces the (f, labels) inputs, ljmds.select() which uses this criterion to pick (h, k).