Struct SemanticClusterAnalyzer 

pub struct SemanticClusterAnalyzer {
    pub points: Vec<ClusterPoint>,
    pub clusters: Vec<Cluster>,
    pub config: AnalyzerConfig,
}

Performs k-means++ style cluster analysis over a set of embedding vectors.

Initialization is deterministic (k-means++ greedy farthest-point selection, no randomness), which means results are reproducible given the same input order.

Fields

points: Vec<ClusterPoint>

All managed points.

clusters: Vec<Cluster>

Clusters produced by the most recent run_kmeans call.

config: AnalyzerConfig

Analyzer configuration.

Implementations

impl SemanticClusterAnalyzer

pub fn new(config: AnalyzerConfig) -> Self

Create a new analyzer with the given configuration.

pub fn add_point(&mut self, id: u64, vector: Vec<f32>)

Add a point to the analyzer.

pub fn run_kmeans(&mut self, k: usize) -> ClusterStats

Run k-means clustering with k clusters.

Initialization (deterministic k-means++ style)

1. First centroid = points[0].vector.
2. Each subsequent centroid = the point whose minimum distance to any already-chosen centroid is the largest (greedy farthest-point).

Iteration

Assign → recompute → check convergence (or max_iterations reached).

Returns a default ClusterStats when k == 0, points is empty, or k > points.len().

pub fn outliers(&self, factor: f64) -> Vec<u64>

Return IDs of points whose distance to their centroid exceeds factor * avg_intra_distance.

If there are no assigned points the list is empty.

pub fn nearest_cluster(&self, query: &[f32]) -> Option<usize>

Return the cluster id of the nearest centroid to query, or None when no clusters exist.

pub fn stats(&self) -> ClusterStats

Return cluster statistics for the current state.

Returns a default ClusterStats when no clusters have been computed.