Struct ComplEx 

pub struct ComplEx {
    pub entity_re: Vec<Vec<f64>>,
    pub entity_im: Vec<Vec<f64>>,
    pub relation_re: Vec<Vec<f64>>,
    pub relation_im: Vec<Vec<f64>>,
    pub dim: usize,
}

ComplEx knowledge graph embedding model.

Entities and relations are embedded in complex vector space ℂ^d. Each embedding is stored as two real vectors (real and imaginary parts).

Score: Re( Σ_k h_k · r_k · conj(t_k) ) = Σ_k ( Re(h)·Re(r)·Re(t) + Im(h)·Re(r)·Im(t) + Re(h)·Im(r)·Im(t) - Im(h)·Im(r)·Re(t) )

Fields

entity_re: Vec<Vec<f64>>

Real part of entity embeddings [n_entities, dim]

entity_im: Vec<Vec<f64>>

Imaginary part of entity embeddings [n_entities, dim]

relation_re: Vec<Vec<f64>>

Real part of relation embeddings [n_relations, dim]

relation_im: Vec<Vec<f64>>

Imaginary part of relation embeddings [n_relations, dim]

dim: usize

Embedding dimension (complex components per embedding)

Implementations

impl ComplEx

pub fn new(n_entities: usize, n_relations: usize, dim: usize) -> Result<Self>

Create a ComplEx model with random initialisation.

pub fn score(&self, h: usize, r: usize, t: usize) -> Result<f64>

Score triple (h, r, t) using the ComplEx scoring function.

pub fn link_prediction_score(&self, h: usize, r: usize, t: usize) -> Result<f64>

Link prediction score.

pub fn predict_tails(&self, h: usize, r: usize, k: usize) -> Result<Vec<usize>>

Return the top-k entity indices most likely to complete (h, r, ?).

pub fn train_epoch(&mut self, dataset: &KGDataset, lr: f64, margin: f64) -> f64

Train one epoch with margin-based negative sampling.

Trait Implementations

impl Clone for ComplEx

fn clone(&self) -> ComplEx

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ComplEx

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

Formats the value using the given formatter.