class RGCNConv(in_channels: Union[int, Tuple[int, int]], out_channels: int, num_relations: int, num_bases: Optional[int] = None, num_blocks: Optional[int] = None, aggr: str = 'mean', root_weight: bool = True, is_sorted: bool = False, bias: bool = True, **kwargs)[source]

Bases: MessagePassing

The relational graph convolutional operator from the “Modeling Relational Data with Graph Convolutional Networks” paper.

\[\mathbf{x}^{\prime}_i = \mathbf{\Theta}_{\textrm{root}} \cdot \mathbf{x}_i + \sum_{r \in \mathcal{R}} \sum_{j \in \mathcal{N}_r(i)} \frac{1}{|\mathcal{N}_r(i)|} \mathbf{\Theta}_r \cdot \mathbf{x}_j,\]

where \(\mathcal{R}\) denotes the set of relations, i.e. edge types. Edge type needs to be a one-dimensional torch.long tensor which stores a relation identifier \(\in \{ 0, \ldots, |\mathcal{R}| - 1\}\) for each edge.


This implementation is as memory-efficient as possible by iterating over each individual relation type. Therefore, it may result in low GPU utilization in case the graph has a large number of relations. As an alternative approach, FastRGCNConv does not iterate over each individual type, but may consume a large amount of memory to compensate. We advise to check out both implementations to see which one fits your needs.


RGCNConv can use dynamic shapes, which means that the shape of the interim tensors can be determined at runtime. If your device doesn’t support dynamic shapes, use FastRGCNConv instead.

  • in_channels (int or tuple) – Size of each input sample. A tuple corresponds to the sizes of source and target dimensionalities. In case no input features are given, this argument should correspond to the number of nodes in your graph.

  • out_channels (int) – Size of each output sample.

  • num_relations (int) – Number of relations.

  • num_bases (int, optional) – If set, this layer will use the basis-decomposition regularization scheme where num_bases denotes the number of bases to use. (default: None)

  • num_blocks (int, optional) – If set, this layer will use the block-diagonal-decomposition regularization scheme where num_blocks denotes the number of blocks to use. (default: None)

  • aggr (str, optional) – The aggregation scheme to use ("add", "mean", "max"). (default: "mean")

  • root_weight (bool, optional) – If set to False, the layer will not add transformed root node features to the output. (default: True)

  • is_sorted (bool, optional) – If set to True, assumes that edge_index is sorted by edge_type. This avoids internal re-sorting of the data and can improve runtime and memory efficiency. (default: False)

  • bias (bool, optional) – If set to False, the layer will not learn an additive bias. (default: True)

  • **kwargs (optional) – Additional arguments of torch_geometric.nn.conv.MessagePassing.

forward(x: Union[Tensor, None, Tuple[Optional[Tensor], Tensor]], edge_index: Union[Tensor, SparseTensor], edge_type: Optional[Tensor] = None)[source]

Runs the forward pass of the module.

  • x (torch.Tensor or tuple, optional) – The input node features. Can be either a [num_nodes, in_channels] node feature matrix, or an optional one-dimensional node index tensor (in which case input features are treated as trainable node embeddings). Furthermore, x can be of type tuple denoting source and destination node features.

  • edge_index (torch.Tensor or SparseTensor) – The edge indices.

  • edge_type (torch.Tensor, optional) – The one-dimensional relation type/index for each edge in edge_index. Should be only None in case edge_index is of type torch_sparse.SparseTensor. (default: None)


Resets all learnable parameters of the module.