Detect outliers from hdbscan for large data

Obtain aggreagted GLOSH outlier scores based on hdbscan

outlier_hdbscan(mat, k, sampleSize, nEpochs, distMethod = "euclidean",
  seed = 1, nproc = 1, distFunc)

Arguments

mat	(numeric matrix) data matrix
k	(pos int) Minimum size of clusters for hdbscan
sampleSize	(pos int) Size of the sample
nEpochs	(pos int) Number of samples
distMethod	(string) Method of compute distance matrix. Default is 'euclidean'
seed	(pos int) seed
nproc	(pos int) Number of parallel processses to use via forking
distFunc	'fun' argument for 'parallelDist::parDist' when distMethod is "custom"

Value

A vector of outlier scores

Examples

set.seed(1)
mix3Gaus <- rbind(
  mvtnorm::rmvnorm(1e3, mean = c(10, 20))
  , mvtnorm::rmvnorm(
    2e3
    , mean = c(20, 30)
    , sigma = matrix(c(1, 0.2, 0.2, 1), ncol = 2))
  , mvtnorm::rmvnorm(100, mean = c(15, 25), sigma = diag(6, 2))
 )
mix3Gaus <- mix3Gaus[sample(nrow(mix3Gaus)), ]

outScore <- outlier_hdbscan(mat = mix3Gaus
                            , k = 100
                            , sampleSize = 1e3
                            , nEpochs    = 1e2
                            )

plot(density(outScore))
plot(mix3Gaus)
plot(mix3Gaus, col = ifelse(outScore > 0.8, 1, 2))