Struct PlaidPruner 

pub struct PlaidPruner {
    pub centroids: Vec<Vec<f32>>,
    /* private fields */
}

PLAID centroid-based candidate pruner.

After train, call candidates at query time to get the set of document IDs whose centroid bag overlaps the query’s centroid bag.

Fields

centroids: Vec<Vec<f32>>

Implementations

impl PlaidPruner

pub fn train( store: &MultiVectorStore, num_centroids: u16, kmeans_iters: usize, seed: u64, ) -> Self

Train the pruner from a MultiVectorStore.

• num_centroids — number of K-means clusters.
• kmeans_iters — Lloyd iterations.
• seed — deterministic seed for centroid initialisation.

pub fn candidates(&self, query: &[Vec<f32>]) -> Vec<u32>

Return candidate doc IDs whose centroid bag overlaps the query’s centroid bag.

The query centroid bag is the set of nearest centroids for each query vector.