Skip to contents

Internal Cluster Validity Indices

Source: R/metrics.R
shoal_metrics.Rd

Computes the Calinski-Harabasz and Davies-Bouldin indices, both of which work from the data rather than a distance matrix, and both of which are used the same way: fit at several values of k and compare.

Usage

shoal_metrics(x, cluster = NULL)

Arguments

x

A numeric matrix or data frame, or a clustering result from this package, in which case both the data and the assignment are taken from it.

cluster

A clustering result or an integer vector of cluster IDs. Required unless x is itself a clustering result. NA entries are excluded.

Value

A one-row data frame with columns n, k, calinski_harabasz and davies_bouldin.

Details

  • Calinski-Harabasz is between-cluster dispersion over within-cluster dispersion, each per degree of freedom. Higher is better.

  • Davies-Bouldin averages, over clusters, the worst-case ratio of combined within-cluster scatter to the distance between centroids. Lower is better.

Both assume roughly convex, centroid-shaped clusters, so they suit shoal_kmeans() and shoal_gmm() better than the density-based algorithms. For arbitrary cluster shapes prefer shoal_silhouette(), which needs only a distance matrix.

See also

shoal_silhouette().

Examples

fit <- shoal_kmeans(as.matrix(iris[, 1:4]), k = 3L)
shoal_metrics(fit)
#>     n k calinski_harabasz davies_bouldin
#> 1 150 3          561.6278      0.6619715

# Comparing candidate values of k
do.call(rbind, lapply(2:5, function(k) {
  shoal_metrics(shoal_kmeans(as.matrix(iris[, 1:4]), k = k))
}))
#>     n k calinski_harabasz davies_bouldin
#> 1 150 2          513.9245      0.4042928
#> 2 150 3          561.6278      0.6619715
#> 3 150 4          530.4871      0.7757009
#> 4 150 5          495.5415      0.8059652