Struct SemanticClusterManager 

pub struct SemanticClusterManager {
    pub clusters: Vec<SemanticCluster>,
    pub config: ClusterManagerConfig,
    pub total_assignments: u64,
}

Online k-means-style semantic cluster manager.

Documents are assigned to the nearest initialised cluster (by Euclidean distance) and each assignment incrementally shifts the cluster centroid toward the document embedding.

Fields

clusters: Vec<SemanticCluster>

The managed clusters (length == config.n_clusters).

config: ClusterManagerConfig

Manager configuration.

total_assignments: u64

Total number of successful assignments since creation or last reset.

Implementations

impl SemanticClusterManager

pub fn new(config: ClusterManagerConfig) -> Self

Create a new manager with n_clusters empty (uninitialized) clusters.

pub fn initialize_centroids(&mut self, centroids: Vec<Vec<f32>>)

Initialise cluster centroids from the provided list.

Only up to min(centroids.len(), n_clusters) clusters are initialised.

pub fn assign( &mut self, doc_id: u64, embedding: &[f32], ) -> Option<ClusterAssignment>

Assign doc_id to the nearest initialised cluster and update the centroid.

A cluster is considered initialised when its centroid is non-empty and its dimension matches that of embedding.

Returns None when no valid cluster exists.

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

Return the cluster_id of the nearest initialised cluster to embedding without mutating any state.

pub fn cluster(&self, cluster_id: usize) -> Option<&SemanticCluster>

Return a reference to a cluster by its cluster_id, or None if out of bounds.

pub fn stats(&self) -> ClusterManagerStats

Compute aggregate statistics over all clusters.

pub fn reset_drift(&mut self)

Reset total_drift to 0.0 for all clusters.