inst/tinytest/test-distance-computation.R

In anticlust: Subset Partitioning via Anticlustering

library("anticlust")

for (M in 1:4) {
  for (K in 2:5) {
    for (i in 3:8) {
      N <- K * i
      features <- matrix(rnorm(N * M), ncol = M)
      distances <- as.matrix(dist(features))
      clusters <- sample(rep(1:K, N/K))
      
      objective <- anticlust:::diversity_objective_(clusters, distances)
      objective2 <- anticlust:::diversity_objective_(clusters, features)
      expect_equal(objective, objective2)
      
      # Swap two items and check objective
      to_swap <- sample(1:K, size = 2)
      swap1 <- which(clusters == to_swap[1])[1]
      swap2 <- which(clusters == to_swap[2])[1]
      
      # Method 1: As in specialized exchange method for anticluster editing
      selected <- anticlust:::selection_matrix_from_clusters(clusters)
      obj1 <- anticlust:::update_objective_distance(distances, selected, swap1, swap2, objective)
      
      # Method 2: As in generic exchange method
      obj2 <- anticlust:::update_objective_generic(distances, clusters, swap1, swap2, diversity_objective)
      expect_equal(obj1, obj2)
      # note, this DOES NOT always work:
      # expect_true(obj1 == obj2) # floating points =/
    }
  }
}