inst/tinytest/test-exchange-method.R

In anticlust: Subset Partitioning via Anticlustering

library("anticlust")

# fast exchange and exchange functions yield the same results - kmeans"
# for K = 2
clusters <- rep(1:2, 5)
features <- matrix(rnorm(20), ncol = 2)
ac <- anticlustering(features, K = clusters, objective = "variance")
ac_fast <- fast_anticlustering(features, clusters, exchange_partners = rep(list(1:10), 10))
expect_equal(all(ac == ac_fast), TRUE)

## for K = 3
clusters <- rep(1:3, 5)
features <- matrix(rnorm(30), ncol = 2)
ac <- anticlustering(features, K = clusters, objective = "variance")
ac_fast <- fast_anticlustering(features, clusters, exchange_partners = rep(list(1:15), 15))
expect_equal(all(ac == ac_fast), TRUE)

## for K = 4
clusters <- rep(1:4, 5)
features <- matrix(rnorm(20), ncol = 1)
ac <- anticlustering(features, K = clusters, objective = "variance")
ac_fast <- fast_anticlustering(features, clusters, exchange_partners = rep(list(1:20), 20))
expect_equal(all(ac == ac_fast), TRUE)


# fast exchange and exchange functions yield the same results - anticluster editing (max diversity)
# for K = 2
K <- 2
clusters <- rep(1:K, 5)
features <- matrix(rnorm(K * 10), ncol = 2)
ac <- anticlustering(features, K = clusters, objective = diversity_objective)
ac_fast <- anticlust:::fast_exchange_dist(as.matrix(dist(features)), clusters, NULL)
expect_equal(all(ac == ac_fast), TRUE)

# for K = 3
K <- 3
clusters <- rep(1:K, 5)
features <- matrix(rnorm(K * 10), ncol = 2)
ac <- anticlustering(features, K = clusters, objective = diversity_objective)
ac_fast <- anticlust:::fast_exchange_dist(as.matrix(dist(features)), clusters, NULL)
expect_equal(all(ac == ac_fast), TRUE)

# for K = 4
K <- 4
clusters <- rep(1:K, 5)
features <- matrix(rnorm(K * 10), ncol = 2)
ac <- anticlustering(features, K = clusters, objective = diversity_objective)
ac_fast <- anticlust:::fast_exchange_dist(as.matrix(dist(features)), clusters, NULL)
expect_equal(all(ac == ac_fast), TRUE)

## Test preclustering restrictions
features <- schaper2019[, 3:6]
clusters <- categorical_sampling(matching(features, 2), K = 2)
ac <- anticlustering(features, K = clusters, objective = diversity_objective)
ac_fast <- anticlustering(features, K = clusters, objective = "distance")
expect_equal(all(ac == ac_fast), TRUE)

## Test categorical restrictions
features <- schaper2019[, 3:6]
ac <- anticlustering(features, K = rep_len(1:2, nrow(schaper2019)),
                     objective = diversity_objective,
                     categories = schaper2019$room)
ac_fast <- anticlustering(features, K = rep_len(1:2, nrow(schaper2019)),
                          objective = "distance",
                          categories = schaper2019$room)
expect_equal(all(ac == ac_fast), TRUE)



# same results for local updating and recomputing - different group sizes"
N <- 60 
M <- 2
clusters <- rep(1:3, c(30, 15, 15))
features <- matrix(rnorm(N * M), ncol = M)
ac1 <- anticlustering(features, K = clusters, objective = "variance")
ac2 <- fast_anticlustering(features, clusters, exchange_partners = rep(list(1:N), N))
ac3 <- anticlustering(features, clusters, objective = "variance")
expect_true(all(ac1 == ac2))
expect_true(all(ac2 == ac3))

ac4 <- anticlustering(features, K = clusters, objective = diversity_objective)
ac5 <- anticlustering(features, clusters)
expect_true(all(ac4 == ac5))