Struct BatchedTrainingLoop 

pub struct BatchedTrainingLoop;

Efficient batched training for knowledge graph completion.

Combines negative-sample generation, batch preparation, and loss computation into a single cohesive API.

Implementations

impl BatchedTrainingLoop

pub fn new() -> Self

Create a new BatchedTrainingLoop.

pub fn prepare_batch( &self, task: &KgCompletionTask, positives: &[Triple], neg_ratio: u32, sampler: &NegativeSampler, ) -> Result<TrainingBatch>

Prepare a TrainingBatch from a slice of positive triples.

For every positive triple, neg_ratio negative triples are sampled via the given sampler.

Returns an error when positives is empty.

pub fn compute_margin_loss( &self, pos_scores: &[f64], neg_scores: &[f64], margin: f64, ) -> Result<f64>

Compute the TransE-style margin ranking loss.

L = Σ_neg max(0, margin − score_pos + score_neg)

Higher scores are assumed to be better for positive triples.

Returns an error when pos_scores or neg_scores is empty.

pub fn compute_binary_cross_entropy( &self, pos_scores: &[f64], neg_scores: &[f64], ) -> Result<f64>

Compute binary cross-entropy loss over positive and negative scores.

L = −(1/N) Σ [log σ(s_pos) + log(1 − σ(s_neg))]

Returns an error when either input slice is empty.

Trait Implementations

impl Clone for BatchedTrainingLoop

fn clone(&self) -> BatchedTrainingLoop

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BatchedTrainingLoop

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

Formats the value using the given formatter.

impl Default for BatchedTrainingLoop

fn default() -> BatchedTrainingLoop

Returns the “default value” for a type.