k-means++ seeding (Geron Ch 9, morie.fn grkmpp) — morie_geron_kmeans_pp

Centre t+1 sampled with P(x) proportional to min_c ||x - c||^2 using the searchsorted-on-cumsum draw, one LCG uniform per centre. Indices are 0-based.

Usage

morie_geron_kmeans_pp_seeding(X, k, seed = 0)

Arguments

X: (m, n) data.
k: Centroids in [1, m].
seed: LCG seed.

Value

List with centroids, indices, min_pairwise_distance, sampling_probabilities.

Examples

set.seed(3)
X <- rbind(matrix(rnorm(20, 0), 10, 2), matrix(rnorm(20, 4), 10, 2))
morie_geron_kmeans_pp_seeding(X, k = 2, seed = 1)
#> $centroids
#>           [,1]      [,2]
#> [1,] 0.1957828 0.1520457
#> [2,] 2.3335252 4.7365021
#> 
#> $indices
#> [1]  4 13
#> 
#> $min_pairwise_distance
#> [1] 5.058378
#> 
#> $sampling_probabilities
#> $sampling_probabilities[[1]]
#>  [1] 0.0064943187 0.0057088201 0.0022956728 0.0055325277 0.0000000000
#>  [6] 0.0007230425 0.0037348275 0.0045071424 0.0095417867 0.0034841921
#> [11] 0.0997927350 0.0916880910 0.1026571588 0.0774833758 0.0703804877
#> [16] 0.0912004378 0.1549071545 0.1159780199 0.0670994808 0.0867907281
#> 
#> 
#> $seed
#> [1] 1
#> 
#> $estimate
#>           [,1]      [,2]
#> [1,] 0.1957828 0.1520457
#> [2,] 2.3335252 4.7365021
#> 
#> $n
#> [1] 20
#>