Writing
Writing and research, newest first.
2026
Why HDBSCAN*’s flat clustering changes with minPts: shrinking clusters, a one-run predictor and a certificate
HDBSCAN* can change its answer when min_samples goes up by one small step: on new benchmark data, with min_cluster_size moving together with min_samples as in the default, about one step in six changed the result a lot. The paper shows why (a selected cluster shrinks below the minimum size), how to predict which cluster is at risk from a single run, and how to prove that a step is safe, at about 1% of the cost of a new run. These are results from a preprint that has not yet been peer reviewed.