Resampling-based clustering: consensus, evidence accumulation, stability

ConsensusClustering: Monti et al. consensus matrix over item subsamples, average-linkage final partition, CDF area and cluster consensus. EvidenceAccumulation: Fred-Jain co-association of many k-means partitions, cut at k or at the largest lifetime. ClusterStability: Hennig's bootstrap Jaccard stability of k-means clusters. All draws are Philox. Identical to the Python arm morie.fn.clusterensemble.

Usage

ConsensusClustering(X, k, n_resamples = 100L, p_item = 0.8, seed = 1)

EvidenceAccumulation(
  X,
  k = NULL,
  n_runs = 50L,
  k_range = c(2, 10),
  seed = 1,
  linkage = "single"
)

ClusterStability(X, k, n_boot = 50L, seed = 1)

Arguments

X

Data matrix (rows are points).

k

Number of clusters (NULL: largest lifetime for EvidenceAccumulation).

n_resamples, p_item

Subsamples and item fraction.

seed

Philox seed.

n_runs, k_range

Partitions and range of their cluster numbers.

linkage

Linkage of the final hierarchy.

n_boot

Bootstrap replicates.

Value

List with cluster.

References

Monti, S., Tamayo, P., Mesirov, J. and Golub, T. (2003). Consensus clustering. Machine Learning 52, 91-118.

Fred, A. L. N. and Jain, A. K. (2005). Combining multiple clusterings using evidence accumulation. IEEE Transactions on Pattern Analysis and Machine Intelligence 27, 835-850.

Hennig, C. (2007). Cluster-wise assessment of cluster stability. Computational Statistics and Data Analysis 52, 258-271.

Examples

X <- rbind(c(0, 0), c(0, 1), c(1, 0), c(8, 8), c(8, 9), c(9, 8))
ConsensusClustering(X, 2, n_resamples = 20)$cluster
#> [1] 1 1 1 2 2 2
EvidenceAccumulation(X, n_runs = 20, k_range = c(2, 4))$cluster
#> [1] 1 1 1 2 2 2