Assign points to clusters by density

Labels every point in the array with a cluster number, or NA when the point is noise. One label per row, in the order you gave the points.

Usage

ga_dbscan(geometry, eps, min_points)

Arguments

geometry: a GeoArrow point array
eps: how close two points must be to be neighbours
min_points: how many neighbours a point needs to seed a cluster

Value

an integer array of cluster labels, the same length as geometry

Details

DBSCAN grows a cluster from any point with at least min_points neighbours within eps. It adds points reachable from that core to the cluster and labels points that never become reachable as noise. Unlike k-means it finds clusters of any shape and does not need the count up front.

ga_dbscan() clusters over the whole array, not within each row, so eps and min_points are single values rather than one per row. It numbers clusters from 1, and the numbers mean nothing beyond grouping. It leaves a row that is not a single point, or is null, out of the clustering and returns null for it.

References

Dbscan

Examples

g <- ga_point(
  c(0, 0.1, 0.2, 5, 5.1, 5.2),
  c(0, 0.1, 0.2, 5, 5.1, 5.2)
)

as.vector(ga_dbscan(g, eps = 1, min_points = 2))
#> [1] 1 1 1 2 2 2