Struct MmrExampleSelector 

pub struct MmrExampleSelector<E> { /* private fields */ }

Maximal Marginal Relevance selector: trades relevance to the input against novelty among already-selected examples. lambda controls the trade-off — 1.0 is pure similarity (equivalent to the semantic selector); 0.0 is pure diversity.

Implementations

impl<E> MmrExampleSelector<E>

where E: Send + Sync + 'static,

pub fn new<F>( embeddings: Arc<dyn Embeddings>, k: usize, lambda: f32, text_of: F, ) -> Self

where F: Fn(&E) -> String + Send + Sync + 'static,

Build with lambda clamped to [0, 1]. k is the number of examples returned.

pub fn with_distance(self, d: Distance) -> Self

Override the distance metric.

Trait Implementations

impl<E> AsyncExampleSelector<E> for MmrExampleSelector<E>

where E: Clone + Send + Sync + 'static,

fn select_async<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, input: &'life1 str, examples: &'life2 [E], ) -> Pin<Box<dyn Future<Output = Result<Vec<E>>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Select examples to include for input.

impl<E> ExampleSelector<E> for MmrExampleSelector<E>

where E: Clone + Send + Sync + 'static,

fn select(&self, input: &str, examples: &[E]) -> Result<Vec<E>>

Select examples from examples to include for input. Returns owned clones (or constructs new examples) so the caller doesn’t have to manage lifetimes against the pool.