Struct ModelInvocationCache 

pub struct ModelInvocationCache { /* private fields */ }

Memoization cache for neural classifier outputs across a single query evaluation. Per impl plan §1.4 decision D-4: cache is scoped per-query (cleared at the start of evaluation); the key is (model_name, ClassifyInput::stable_hash).

Eviction policy: a naive “clear-when-full” heuristic — when the cache reaches max_entries, the entire map is dropped. This keeps the type allocation-light and avoids dragging in an LRU dep for v1. Documented in impl plan as Stage 1 trade-off; a proper LRU follow-up can swap the inner type without changing the public API.

Implementations

impl ModelInvocationCache

pub fn new(max_entries: usize) -> Self

pub fn get(&self, model: &str, input_hash: u64) -> Option<f64>

Lookup. Returns Some(prob) on hit, None on miss.

pub fn insert(&self, model: &str, input_hash: u64, value: f64)

Insert. On overflow (cache size ≥ max_entries), drops the entire cache before inserting — naive but bounded. Callers should size max_entries for the expected working set.

pub fn clear(&self)

Empty the cache. Useful for LocyConfig users who reuse a shared cache across evaluations and want explicit reset.

pub fn len(&self) -> usize

pub fn is_empty(&self) -> bool

Trait Implementations

impl Debug for ModelInvocationCache

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ModelInvocationCache

fn default() -> ModelInvocationCache

Returns the “default value” for a type.