Get the consensus NMF stability diagnostics — get

Returns the mean silhouette of the consensus clusters, the relative reconstruction errors and the per-component clustering table. A stability near 1 means every restart found the same programmes; a low one means the factorisation is not reproducible at this k.

Usage

get_stability(x)

# S3 method for class 'ConsensusNmfResult'
get_stability(x)

Arguments

x: An object holding consensus NMF results.

Value

A list with stability, rel_error, rel_run_errors, clusters, cluster_sizes, n_dropped and n_empty_clusters.

Examples

# how much the restarts agreed on the consensus factors
sc <- demo_single_cells()
res <- consensus_nmf_sc(
  sc,
  k = 5L,
  n_runs = 5L,
  nmf_consensus_params = params_nmf_consensus(density_threshold = 2),
  .verbose = FALSE
)
get_stability(res)$stability
#> [1] 0.9446008

unlink(sc@dir_data, recursive = TRUE, force = TRUE)