Struct UfpIndex 

pub struct UfpIndex { /* private fields */ }

Index structure with lock-free concurrent access via DashMap and ANN support

Implementations

impl UfpIndex

Provides semantic & perceptual retrieval methods

pub fn search( &self, query: &IndexRecord, mode: QueryMode, top_k: usize, ) -> Result<Vec<QueryResult>, IndexError>

Search for top-k most similar entries.

impl UfpIndex

pub fn new(cfg: IndexConfig) -> Result<Self, IndexError>

Initialize or open an index using the configured backend. This will build the backend from the config.

pub fn with_backend(cfg: IndexConfig, backend: Box<dyn IndexBackend>) -> Self

Build an index with a custom backend (e.g., in-memory for tests). This is useful for dependency injection and testing.

pub fn semantic_vector_count(&self) -> usize

Get the number of vectors in the semantic index.

pub fn should_use_ann(&self) -> bool

Check if ANN search should be used for the current dataset size.

pub fn rebuild_ann_if_needed(&self)

Rebuild the ANN index if needed. This should be called periodically after batch insertions.

pub fn quantize(vec: &Array1<f32>, scale: f32) -> QuantizedVec

Quantize float embeddings -> i8 using a raw scale. This is a simple linear quantization with clamping.

pub fn quantize_with_strategy( vec: &Array1<f32>, cfg: &QuantizationConfig, ) -> QuantizedVec

Quantize using a configured strategy. This allows for different quantization strategies to be used in the future.

pub fn upsert(&self, rec: &IndexRecord) -> Result<(), IndexError>

Insert or update a record. The record is encoded and compressed before being sent to the backend.

pub fn delete(&self, hash: &str) -> Result<(), IndexError>

Remove a record from the index.

pub fn flush(&self) -> Result<(), IndexError>

Flush backend buffers if supported. This is useful for ensuring data is written to disk.

pub fn get(&self, hash: &str) -> Result<Option<IndexRecord>, IndexError>

Retrieve a record by hash. The record is decompressed and decoded after being retrieved from the backend.

pub fn scan( &self, visitor: &mut dyn FnMut(&IndexRecord) -> Result<(), IndexError>, ) -> Result<(), IndexError>

Scan all records in the index. Iterates over all records in the backend, decodes them, and passes them to the visitor.

pub fn batch_insert(&self, records: &[IndexRecord]) -> Result<(), IndexError>

Batch insert multiple records (efficient for large datasets). This can be much faster than calling upsert in a loop.