spore-clustering 0.1.0b2

The SPORE clustering algorithm.

SPORE

Skeleton Propagation Over Recalibrating Expansions — a graph-based clustering algorithm for arbitrary-shape, arbitrary-scale clusters.

How it works

SPORE builds clusters in three stages:

1. k-NN graph construction — a global nearest-neighbor graph is built (exact or approximate), with neighbor counts scaling as ~O(log N) by default.
2. Variance-aware BFS expansion — clusters are seeded from densest points outward. Edges are accepted only if their distance is statistically consistent with the cluster's evolving internal distance distribution — mean and variance updated incrementally as expansion proceeds. This prevents bridging across low-density gaps while preserving irregular shapes.
3. Reassignment — clusters below min_cluster_size are merged into nearby larger ones using a composite score weighing proximity, relative size, density, and angular isotropy, or labeled as noise.

Installation

pip install spore-clustering

Quick start

from spore_clustering import SPORE

labels = SPORE().fit_predict(X)

Key parameters

Parameter	Description
expansion	Z-score threshold controlling how aggressively clusters grow
neighborhood_percentile	Bounded alternative to expansion; typical values: 25, 50, 75, 93.75
retention_rate	Fraction of neighbors that must pass variance filter to continue expansion
min_cluster_size	Minimum cluster size (int) or exponent for N-relative scaling (float)

See the full API reference for all parameters.

Reusing a precomputed neighbor index

dindex = SPORE.DataIndex(
    connectivity=k,
    neighbors=neighbors,
    dists=distances,
    dataset_scale=scale,
)

labels = SPORE(dindex=dindex, retention_rate=0.25).fit_predict(X)

Complexity

With approximate k-NN and default neighbor scaling (k ~ log N):

Phase	Complexity
k-NN construction	O(Nd log N)
BFS expansion	O(N log N)
Reassignment	O(Rd log N), R ≤ N

scikit-learn compatibility

SPORE follows standard scikit-learn estimator conventions: fit, fit_predict, get_params, set_params.