InductiveNodePieceGNN

class InductiveNodePieceGNN(*, gnn_encoder: Iterable[Module] | None = None, **kwargs)[source]

Bases: InductiveNodePiece

Inductive NodePiece with a GNN encoder on top.

Overall, it’s a 3-step procedure:

  1. Featurizing nodes via NodePiece

  2. Message passing over the active graph using NodePiece features

  3. Scoring function for a given batch of triples

As of now, message passing is expected to be over the full graph

Initialize the model.

Parameters:

  • gnn_encoder (Iterable[Module] | None) – an iterable of message passing layers. Defaults to 2-layer CompGCN with Hadamard composition.

  • kwargs – additional keyword-based parameters passed to InductiveNodePiece.__init__.

Methods Summary

reset_parameters_()

Reset the GNN encoder explicitly in addition to other params.

Methods Documentation

reset_parameters_()[source]

Reset the GNN encoder explicitly in addition to other params.