Enum UniversalPoolingStrategy 

pub enum UniversalPoolingStrategy {
    ClsToken,
    Mean,
    Max,
    MeanSqrt,
    AttentionPooling,
    WeightedMean,
}

Pooling strategy for UniversalSentenceEncoder.

Each variant describes how per-token embeddings are collapsed into a single fixed-length sentence vector.

Variants

ClsToken

Use only the embedding of the first token (CLS-token style).

Mean

Arithmetic mean over all token embeddings.

Max

Component-wise maximum over all token embeddings.

MeanSqrt

Mean divided by √(n_tokens).

This down-scales the pooled vector for longer sequences, which can help when sequences have variable lengths.

AttentionPooling

Attention-weighted mean with a learnable query vector q.

Scores are computed as softmax(E·q), then the result is Eᵀ·scores. Requires calling UniversalSentenceEncoder::fit_attention_pooling before use, or the mean pool is used as a fallback.

WeightedMean

Weighted mean using log-IDF weights per token.

Requires calling UniversalSentenceEncoder::fit_idf_weights before use, or the mean pool is used as a fallback.

Trait Implementations

impl Clone for UniversalPoolingStrategy

fn clone(&self) -> UniversalPoolingStrategy

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for UniversalPoolingStrategy

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

Formats the value using the given formatter.

impl PartialEq for UniversalPoolingStrategy

fn eq(&self, other: &UniversalPoolingStrategy) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for UniversalPoolingStrategy