Choosing k

k defines the current extraction. It must satisfy 2 <= k <= k_max.

clusterer.fit(X, k=10)

Interpretation

Larger k values generally smooth the density estimate more strongly. Smaller k values make the hierarchy more sensitive to local structure.

Core-SG makes it cheap to compare several valid k values after the initial build, so a practical workflow is to fit once at the largest candidate and inspect labels, probabilities, persistence, and tree artifacts across the candidate list.