Skip to contents

Seed to initialize farthest-first selection

Source: R/seed.R
PickPoint.Rd

PickPoint() implements a range of strategies to select a seed for greedy farthest-first selection. Propitious seeds yield better solutions.

Usage

PickPoint(
  d = NULL,
  points = NULL,
  strategy = c("peripheral", "anti_centroid", "random_furthest", "diameter",
    "anti_medoid", "medoid", "rowsum", "rownorm")
)

Arguments

d

A dist object or square symmetric numeric matrix. Ignored when points is supplied.

points

Optional N x dim numeric coordinate matrix; when supplied the seed is computed from coordinates in O(N) memory. Required for the "anti_centroid" anchor, which has no distance-matrix form.

strategy

Character specifying method to employ:

"peripheral"(default)

Two sweeps: the point furthest from point 1, then the point furthest from that (a diameter-endpoint approximation). \(O(N)\).

"anti_centroid"

The point farthest from the coordinate mean (\(\argmax ||x - x_bar||\)). \(O(N * dim)\). Requires points.

"random_furthest"

The point furthest from a random pivot. \(O(N)\).

"diameter"

A row endpoint of the diameter pair (the maximum pairwise distance).

"medoid"

The 1-median (medoid): the point minimising the sum of distances to all others.

"anti_medoid"

The point furthest from the 1-median (medoid).

"rowsum"

The point maximising the sum of distances to all others (the 1-anti-median).

"rownorm"

The point maximising \(\sqrt(\sum{d^2})\), the L2 counterpart of "rowsum".

Value

PickPoint() returns an integer that identifies the index of a proposed seed in d or points.

See also

FarFirst(), which seeds and runs the greedy pass in one call.

Examples

set.seed(1)
pts <- matrix(rnorm(60), ncol = 2)
d <- dist(pts)
PickPoint(d, strategy = "diameter")
#> [1] 14
FarFirst(5L, d, strategy = PickPoint(d, strategy = "diameter"))
#> 5 elements (14 4 26 5 28) selected by farthest-first, each at distance >= 1.765