NEWS

gmeans 0.2.0.9000

gmeans() now splits a cluster by starting the two new centers along its main principal component, as described in Hamerly and Elkan (2003), instead of at random points. The split step no longer depends on the random seed and the number of clusters is found more reliably, so results may differ from earlier versions.
gmeans() now uses k_init = 1 by default, as in Hamerly and Elkan (2003). The previous default of 2 could never return a single cluster. With one initial center the result no longer depends on the random seed.
gmeans() now uses level = 0.0001 by default, the significance level used by Hamerly and Elkan (2003). The previous default of 0.05 often split clusters that are Gaussian, especially in higher dimensions.
gmeans() now uses k_max = Inf by default, as in Hamerly and Elkan (2003). The previous default of 10 silently capped the number of clusters on data with more than 10 clusters.
gmeans() gains an iter.max argument that defaults to 100, instead of the 10 used by stats::kmeans(), so k-means converges when there are many clusters instead of warning that it did not converge.
gmeans() now stores k_init, k_max, and level in the returned object.
gmeans() now errors clearly when centers is passed, pointing to k_init instead.
gmeans() now errors clearly when k_init is not less than the number of rows in x.
New summary() method for gmeans objects.
New tidy(), augment(), and glance() methods for gmeans objects.

Removed the mlr3 integration vignette since the learner now ships in mlr3cluster as lrn("clust.gmeans").
gmeans() now errors when k_init exceeds k_max.
gmeans() now requires finite numeric input (logical is coerced to 0/1) and errors clearly when there are fewer distinct points than centers.
gmeans() now errors clearly on input with zero rows or zero columns.
predict.gmeans() now errors when the Minkowski power p is not positive.
predict.gmeans() and compute_wss() now accept data frames with unused non-numeric columns and error clearly when newdata is not a matrix or data frame.
predict.gmeans() and compute_wss() now error on a column mismatch between newdata and unnamed centers instead of returning wrong distances.