RGCN

class RGCN(*, triples_factory: ~pykeen.triples.triples_factory.CoreTriplesFactory, embedding_dim: int = 500, num_layers: int = 2, base_entity_initializer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function xavier_uniform_>, base_entity_initializer_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, relation_representations: str | ~pykeen.nn.representation.Representation | type[~pykeen.nn.representation.Representation] | None = None, relation_initializer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function xavier_uniform_>, relation_initializer_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, interaction: str | ~pykeen.nn.modules.Interaction[~torch.Tensor, ~pykeen.typing.RelationRepresentation, ~torch.Tensor] | type[~pykeen.nn.modules.Interaction[~torch.Tensor, ~pykeen.typing.RelationRepresentation, ~torch.Tensor]] | None = 'DistMult', interaction_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, use_bias: bool = True, activation: str | ~torch.nn.modules.module.Module | None = None, activation_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, edge_dropout: float = 0.4, self_loop_dropout: float = 0.2, edge_weighting: str | ~pykeen.nn.weighting.EdgeWeighting | None = None, decomposition: str | ~pykeen.nn.message_passing.Decomposition | None = None, decomposition_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, regularizer: str | ~pykeen.regularizers.Regularizer | None = None, regularizer_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, **kwargs)[source]

Bases: ERModel[Tensor, RelationRepresentation, Tensor]

An implementation of R-GCN from [schlichtkrull2018].

The Relational Graph Convolutional Network (R-GCN) comprises three parts:

A GCN-based entity encoder that computes enriched representations for entities, cf. pykeen.nn.message_passing.RGCNRepresentations. The representation for entity \(i\) at level \(l \in (1,\dots,L)\) is denoted as \(\textbf{e}_i^l\). The GCN is modified to use different weights depending on the type of the relation.
Relation representations \(\textbf{R}_{r} \in \mathbb{R}^{d \times d}\) is a diagonal matrix that are learned independently from the GCN-based encoder.
An arbitrary interaction model which computes the plausibility of facts given the enriched representations, cf. pykeen.nn.modules.Interaction.

Scores for each triple \((h,r,t) \in \mathcal{K}\) are calculated by using the representations in the final level of the GCN-based encoder \(\textbf{e}_h^L\) and \(\textbf{e}_t^L\) along with relation representation \(\textbf{R}_{r}\). While the original implementation of R-GCN used the DistMult model and we use it as a default, this implementation allows the specification of an arbitrary interaction model.

\[f(h,r,t) = \textbf{e}_h^L \textbf{R}_{r} \textbf{e}_t^L\]