Open Access

Table 2.

Test results on simulated cluster samples.

Open cluster sample (N = 12) (a)
Compact cluster sample (N = 37) (b)
SigMA DBSCAN HDBSCAN SigMA DBSCAN HDBSCAN
NMI 0.95 0.94 0.86 0.60 ±0.02 0.39 ± 0.06 0.33 ± 0.16
AMI 0.95 0.94 0.86 0.56 ±0.01 0.38 ± 0.06 0.32 ± 0.16
ARI 0.96 0.95 0.74 0.38 ±0.03 0.11 ± 0.03 0.08 ± 0.06
Precision 0.97 0.98 0.74 0.64 ±0.05 0.23 ± 0.05 0.22 ± 0.15
Recall 0.98 0.96 0.76 0.47 ±0.04 0.25 ± 0.03 0.20 ± 0.06
Contamination 0.03 0.02 0.08 0.24 ±0.13 0.13 ±0.12 0.25 ± 0.22
Completeness 0.98 0.96 0.92 0.76 ±0.15 0.79 ± 0.24 0.76 ± 0.21
ACC 0.97 0.96 0.76 0.47 ±0.04 0.25 ± 0.03 0.20 ± 0.06
BACC 0.90 0.88 0.88 0.51 ±0.04 0.16 ± 0.03 0.18 ± 0.07
MC 0.96 0.95 0.76 0.55 ±0.04 0.23 ± 0.04 0.16 ± 0.11
Ntot 12 12 14 26.8 ±2.0 10.0 ± 1.7 24.2 ± 4.8
Ncluster 12 12 12 24.1 ±1.7 10.0 ± 1.7 12.0 ± 5.5

Notes. Bold-faced numbers indicate the best performance given a specific evaluation metric. The three clustering methods, SigMA, DBSCAN, and HDBSCAN, are applied to two data sets.

(a)

The open cluster sample contains 12 clusters. It is a subset of the Gaia mock EDR3 catalog (Rybizki et al. 2020).

(b)

The compact cluster samples contain 37 clusters. It mimics the cluster environment of Sco-Cen, where groups are densely packed together. The 10 compact cluster samples are generated in a random effects model, and the resulting distribution varies substantially across realizations. To estimate the performance of clustering algorithms on the compact cluster sample, we average performances across ten individual samples. We report the mean and standard deviation of performance scores.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.