Struct BatchSemanticClusterManager 

pub struct BatchSemanticClusterManager { /* private fields */ }

Batch k-means semantic cluster manager.

Runs Lloyd’s k-means algorithm over a set of named embeddings, then supports prediction (nearest centroid lookup), membership queries, and quality metrics.

Implementations

impl BatchSemanticClusterManager

pub fn new(config: BatchClusterConfig) -> Self

Create a new (unfitted) batch cluster manager.

pub fn fit(&mut self, embeddings: &[(String, Vec<f64>)]) -> Result<(), String>

Run k-means clustering on the provided embeddings.

Embeddings are (doc_id, vector) pairs. The method initialises centroids from the first k distinct embeddings and then iterates assignment + update steps until convergence or max_iterations is reached.

Calling fit again resets all previous state.

pub fn predict(&self, embedding: &[f64]) -> Result<usize, String>

Predict which cluster a new embedding belongs to (nearest centroid).

Returns an error if the manager has not been fitted yet.

pub fn get_cluster(&self, id: usize) -> Option<&BatchCluster>

Return a reference to a cluster by id, or None if out of range.

pub fn get_assignment(&self, doc_id: &str) -> Option<usize>

Return the cluster assignment for a document, or None if unknown.

pub fn cluster_members(&self, cluster_id: usize) -> Vec<String>

Return the list of document IDs assigned to a given cluster.

pub fn total_inertia(&self) -> f64

Total inertia (sum of squared distances) across all clusters.

pub fn silhouette_score_approx( &self, embeddings: &[(String, Vec<f64>)], ) -> Result<f64, String>

Approximate silhouette score.

For each point, computes (b - a) / max(a, b) where:

• a = average Euclidean distance to other members of the same cluster
• b = average distance to the nearest other cluster centroid

Returns the mean silhouette across all points.

pub fn stats(&self) -> BatchClusterManagerStats

Return aggregate statistics about the current clustering state.