Calculate DBCV Metric — dbcv • poem

Compute the DBCV (Density-Based Clustering Validation) metric.

Usage

dbcv(
  X,
  labels,
  distance = "euclidean",
  noise_id = -1,
  check_duplicates = FALSE,
  use_igraph_mst = TRUE,
  BPPARAM = BiocParallel::SerialParam(),
  ...
)

Arguments

X: Numeric matrix of samples.
labels: Integer vector of cluster IDs.
distance: String specifying the distance metric. "sqeuclidean", or possible method in stats::dist(). By default "euclidean".
noise_id: Integer, the cluster ID in y for noise (default -1).
check_duplicates: Logical flag to check for duplicate samples.
use_igraph_mst: Logical flag to use igraph's MST implementation. Currently only mst from igraph is implemented.
BPPARAM: BiocParallel params for multithreading (default none)
...: Ignored

Value

A list:

vcs: Numeric vector of validity index for each cluster.
dbcv: Numeric value representing the overall DBCV metric.

Details

This implementation will not fully reproduce the results of other existing implementations (e.g. https://github.com/FelSiq/DBCV) due to the different algorithms used for computing the Minimum Spanning Tree.

References

Davoud Moulavi, et al. 2014; 10.1137/1.9781611973440.96.

Examples

data(noisy_moon)
data <- noisy_moon
dbcv(data[, c("x", "y")], data$kmeans_label)
#> $vcs
#> [1] -0.4383721 -0.4077112
#> 
#> $dbcv
#> [1] -0.4230416
#> 
dbcv(data[, c("x", "y")], data$hdbscan_label)
#> $vcs
#>  [1] -0.5889023  0.3726825  0.5500422  0.7884686  0.4887283  0.7682203
#>  [7]  0.7246492  0.7246492  0.9349664  0.4749650  0.5500422  0.3726825
#> 
#> $dbcv
#> [1] 0.4214685
#>