Struct Builder 

pub struct Builder { /* private fields */ }

Builds one index in three steps: create, then select_representatives or select_representatives_custom, then build. One build per Builder.

Implementations

impl Builder

pub fn new( store: ReadableWritableListableStorage, levels: u32, metric: Metric, is_normalized: bool, memory_limit_bytes: usize, embedding_dtype: Option<EmbeddingDtype>, chunks: ChunkSizes, ) -> Result<Self>

Starts a build in store instead of a path. Otherwise the same as Self::create.

pub fn create( index_path: &Path, levels: u32, metric: Metric, is_normalized: bool, memory_limit_bytes: usize, embedding_dtype: Option<EmbeddingDtype>, chunks: ChunkSizes, ) -> Result<Self>

Creates an index at index_path. levels is the number of node levels below the root. Set is_normalized only if every embedding is unit-length, and leave embedding_dtype as None to keep each source’s own dtype.

pub fn select_representatives( &mut self, source: &EmbeddingsSource, target_cluster_items: usize, strategy: RepresentativeStrategy, fallback_batch_vecs: usize, ) -> Result<()>

Picks the representatives the tree is built from, chosen by strategy. target_cluster_items is the average cluster size it aims for, and fallback_batch_vecs is the chunk size assumed when source isn’t chunked.

pub fn select_representatives_custom( &mut self, ids: Array1<u32>, embeddings: Array2<f32>, ) -> Result<()>

Uses caller-chosen representatives instead of picking them, such as ones from an external clustering step. ids[i] is the id of row i of embeddings.

pub fn build( &mut self, dataset: &EmbeddingsSource, fallback_batch_vecs: usize, ) -> Result<()>

Builds the tree from dataset, whose rows get item ids 0, 1, 2, … in order. fallback_batch_vecs works as in select_representatives.