Struct EmbeddingSimilarityCache 

pub struct EmbeddingSimilarityCache { /* private fields */ }

Two-level cache for cosine similarity scores between embedding ID pairs.

Entries are evicted using a Least-Frequently-Used (LFU) policy when the cache reaches max_capacity, and are considered stale after ttl_secs seconds.

Implementations

impl EmbeddingSimilarityCache

pub fn new(max_capacity: usize, ttl_secs: u64) -> Self

Create a new cache with the given capacity and TTL.

pub fn get(&mut self, a: u64, b: u64) -> Option<f32>

Look up the similarity score for the pair (a, b).

• Returns Some(score) and increments hit_count + stats.hits on a fresh hit.
• Removes the entry and increments stats.misses when stale.
• Returns None and increments stats.misses when absent.

pub fn insert(&mut self, a: u64, b: u64, score: f32)

Insert a similarity score for the pair (a, b).

When the cache is at capacity the entry with the lowest hit_count is evicted first (ties broken arbitrarily).

pub fn cosine_similarity(a: &[f32], b: &[f32]) -> f32

Compute the cosine similarity between two dense vectors.

Returns 0.0 if either vector has a zero norm. Vectors of different lengths are treated as though padded with zeros — only the shorter length is used for the dot product, and each norm is computed over its own full slice.

pub fn compute_and_cache( &mut self, a_id: u64, a_vec: &[f32], b_id: u64, b_vec: &[f32], ) -> f32

Return the cached similarity for (a_id, b_id), computing and caching it on a miss.

pub fn evict_stale(&mut self) -> usize

Remove all entries whose TTL has expired.

Returns the number of entries removed and updates stats.evictions.

pub fn stats(&self) -> &CacheStats

Return a reference to the current cache statistics.

pub fn clear(&mut self)

Clear all entries and reset all statistics.