|
Clustering
C++20 header-only: DBSCAN, HDBSCAN, k-means.
|
Tuning knobs for the NN-Descent MST backend. More...
#include <clustering/hdbscan/policy/nn_descent_mst_backend.h>
Public Attributes | |
| std::size_t | kExtra = 10 |
Extra neighbours per node on top of minSamples; the kNN graph is built with k = minSamples + kExtra. | |
| std::size_t | maxIter = 10 |
Iteration cap on the NN-Descent join loop; forwarded to NnDescentIndex. | |
| float | delta = 0.001F |
Convergence threshold on the update fraction; forwarded to NnDescentIndex. | |
| std::uint64_t | seed = 0 |
PRNG seed for RP-tree partition choices; forwarded to NnDescentIndex. | |
Tuning knobs for the NN-Descent MST backend.
kExtra widens the kNN-graph k beyond minSamples so the MRD edge set carries enough candidates for Kruskal to recover an MST close to the exact Prim result. maxIter, delta, and seed forward directly to NnDescentIndex.
Definition at line 30 of file nn_descent_mst_backend.h.
| float clustering::hdbscan::NnDescentMstConfig::delta = 0.001F |
Convergence threshold on the update fraction; forwarded to NnDescentIndex.
Definition at line 40 of file nn_descent_mst_backend.h.
| std::size_t clustering::hdbscan::NnDescentMstConfig::kExtra = 10 |
Extra neighbours per node on top of minSamples; the kNN graph is built with k = minSamples + kExtra.
Sized for Dong 2011's k = 15 recall target at default min_samples = 5. Larger values trade build time for MRD-edge coverage; raise this when the input is near-uniform and the MRD-MST topology depends on edges beyond each point's 15 nearest.
Definition at line 36 of file nn_descent_mst_backend.h.
| std::size_t clustering::hdbscan::NnDescentMstConfig::maxIter = 10 |
Iteration cap on the NN-Descent join loop; forwarded to NnDescentIndex.
Definition at line 38 of file nn_descent_mst_backend.h.
| std::uint64_t clustering::hdbscan::NnDescentMstConfig::seed = 0 |
PRNG seed for RP-tree partition choices; forwarded to NnDescentIndex.
Definition at line 42 of file nn_descent_mst_backend.h.