Semi-supervised learning via k-means representative labeling

Clusters the unlabeled pool (Lloyd/k-means++ on a deterministic LCG stream), finds each cluster's real-point representative closest to its centroid, labels that representative from the nearest labeled point, and propagates the label to every cluster member.

Usage

morie_geron_semisupervised_cluster(
  X,
  X_labeled,
  y_labeled,
  n_clusters = 2,
  seed = 0,
  y_true = NULL
)

Arguments

X

Unlabeled pool (n, d).

X_labeled

Labeled instances (m, d).

y_labeled

Their labels, length m.

n_clusters

Labeling budget.

seed

LCG seed.

y_true

Optional gold labels to score against.

Value

list with labels, cluster, centers, representatives, representative_labels, accuracy, estimate, n, method.

Examples

morie_geron_semisupervised_cluster(X = c(1, 2, 3, 4, 5, 6, 7, 8),
  X_labeled = c(1, 2, 3, 4, 5, 6, 7, 8), y_labeled = c(1, 2, 3, 4, 5, 6, 7, 8))
#> $labels
#> [1] 2 2 2 2 6 6 6 6
#> 
#> $cluster
#> [1] 0 0 0 0 1 1 1 1
#> 
#> $centers
#>      [,1]
#> [1,]  2.5
#> [2,]  6.5
#> 
#> $representatives
#> [1] 1 5
#> 
#> $representative_labels
#> [1] 2 6
#> 
#> $accuracy
#> NULL
#> 
#> $estimate
#> [1] 0.25
#> 
#> $n
#> [1] 8
#> 
#> $method
#> [1] "k-means clustering, nearest-labeled representative labeling, propagation to members"
#>