Struct NmeScClusterer 

pub struct NmeScClusterer { /* private fields */ }

NME-SC (Normalized Maximum Eigengap Spectral Clustering) clusterer.

Thin adapter over the shared spectral graph (k-NN affinity + Laplacian spectrum in crate::spectral): the normalized-maximum eigengap picks k directly, then kmeans_pp runs on the spectral embedding.

Implementations

impl NmeScClusterer

pub fn new(max_clusters: usize) -> Self

{ true } pub fn new(max_clusters: usize) -> Self { ret.max_clusters >= 1 }

Trait Implementations

impl Clusterer for NmeScClusterer

Available on crate feature spectral only.

fn cluster(&self, embeddings: &[Vec<f32>]) -> Result<Vec<usize>, ClustererError>

Cluster embeddings. Each inner vector must have the same length and be approximately L2-normalized.

fn max_clusters(&self) -> usize

Hard ceiling on the number of clusters this implementation can produce.

fn cluster_with_durations( &self, embeddings: &[Vec<f32>], durations_secs: &[f64], ) -> Result<Vec<usize>, ClustererError>

Cluster with per-embedding durations (seconds). Defaults to ignoring durations and calling Self::cluster. Backends that filter short embeddings (cVBx short-segment exclusion) override this so unreliable short windows are kept out of AHC/VB and reassigned afterward.

fn wants_raw_embeddings(&self) -> bool

Whether this clusterer wants raw (non-L2-normalized) embeddings. Cosine clusterers (AHC, NME-SC) are scale-invariant and default to false; a PLDA backend needs the original embedding scale for its mean-centering and overrides to true.

impl Default for NmeScClusterer

Available on crate feature spectral only.

fn default() -> Self

Returns the “default value” for a type.