class RGATConv(in_channels: int, out_channels: int, num_relations: int, num_bases: Optional[int] = None, num_blocks: Optional[int] = None, mod: Optional[str] = None, attention_mechanism: str = 'across-relation', attention_mode: str = 'additive-self-attention', heads: int = 1, dim: int = 1, concat: bool = True, negative_slope: float = 0.2, dropout: float = 0.0, edge_dim: Optional[int] = None, bias: bool = True, **kwargs)[source]

Bases: MessagePassing

The relational graph attentional operator from the “Relational Graph Attention Networks” paper.

Here, attention logits \(\mathbf{a}^{(r)}_{i,j}\) are computed for each relation type \(r\) with the help of both query and key kernels, i.e.

\[\mathbf{q}^{(r)}_i = \mathbf{W}_1^{(r)}\mathbf{x}_{i} \cdot \mathbf{Q}^{(r)} \quad \textrm{and} \quad \mathbf{k}^{(r)}_i = \mathbf{W}_1^{(r)}\mathbf{x}_{i} \cdot \mathbf{K}^{(r)}.\]

Two schemes have been proposed to compute attention logits \(\mathbf{a}^{(r)}_{i,j}\) for each relation type \(r\):

Additive attention

\[\mathbf{a}^{(r)}_{i,j} = \mathrm{LeakyReLU}(\mathbf{q}^{(r)}_i + \mathbf{k}^{(r)}_j)\]

or multiplicative attention

\[\mathbf{a}^{(r)}_{i,j} = \mathbf{q}^{(r)}_i \cdot \mathbf{k}^{(r)}_j.\]

If the graph has multi-dimensional edge features \(\mathbf{e}^{(r)}_{i,j}\), the attention logits \(\mathbf{a}^{(r)}_{i,j}\) for each relation type \(r\) are computed as

\[\mathbf{a}^{(r)}_{i,j} = \mathrm{LeakyReLU}(\mathbf{q}^{(r)}_i + \mathbf{k}^{(r)}_j + \mathbf{W}_2^{(r)}\mathbf{e}^{(r)}_{i,j})\]


\[\mathbf{a}^{(r)}_{i,j} = \mathbf{q}^{(r)}_i \cdot \mathbf{k}^{(r)}_j \cdot \mathbf{W}_2^{(r)} \mathbf{e}^{(r)}_{i,j},\]

respectively. The attention coefficients \(\alpha^{(r)}_{i,j}\) for each relation type \(r\) are then obtained via two different attention mechanisms: The within-relation attention mechanism

\[\alpha^{(r)}_{i,j} = \frac{\exp(\mathbf{a}^{(r)}_{i,j})} {\sum_{k \in \mathcal{N}_r(i)} \exp(\mathbf{a}^{(r)}_{i,k})}\]

or the across-relation attention mechanism

\[\alpha^{(r)}_{i,j} = \frac{\exp(\mathbf{a}^{(r)}_{i,j})} {\sum_{r^{\prime} \in \mathcal{R}} \sum_{k \in \mathcal{N}_{r^{\prime}}(i)} \exp(\mathbf{a}^{(r^{\prime})}_{i,k})}\]

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.

To enhance the discriminative power of attention-based GNNs, this layer further implements four different cardinality preservation options as proposed in the “Improving Attention Mechanism in Graph Neural Networks via Cardinality Preservation” paper:

\[ \begin{align}\begin{aligned}\text{additive:}~~~\mathbf{x}^{{\prime}(r)}_i &= \sum_{j \in \mathcal{N}_r(i)} \alpha^{(r)}_{i,j} \mathbf{x}^{(r)}_j + \mathcal{W} \odot \sum_{j \in \mathcal{N}_r(i)} \mathbf{x}^{(r)}_j\\\text{scaled:}~~~\mathbf{x}^{{\prime}(r)}_i &= \psi(|\mathcal{N}_r(i)|) \odot \sum_{j \in \mathcal{N}_r(i)} \alpha^{(r)}_{i,j} \mathbf{x}^{(r)}_j\\\text{f-additive:}~~~\mathbf{x}^{{\prime}(r)}_i &= \sum_{j \in \mathcal{N}_r(i)} (\alpha^{(r)}_{i,j} + 1) \cdot \mathbf{x}^{(r)}_j\\\text{f-scaled:}~~~\mathbf{x}^{{\prime}(r)}_i &= |\mathcal{N}_r(i)| \odot \sum_{j \in \mathcal{N}_r(i)} \alpha^{(r)}_{i,j} \mathbf{x}^{(r)}_j\end{aligned}\end{align} \]
  • If attention_mode="additive-self-attention" and concat=True, the layer outputs heads * out_channels features for each node.

  • If attention_mode="multiplicative-self-attention" and concat=True, the layer outputs heads * dim * out_channels features for each node.

  • If attention_mode="additive-self-attention" and concat=False, the layer outputs out_channels features for each node.

  • If attention_mode="multiplicative-self-attention" and concat=False, the layer outputs dim * out_channels features for each node.

Please make sure to set the in_channels argument of the next layer accordingly if more than one instance of this layer is used.


For an example of using RGATConv, see examples/

  • in_channels (int) – Size of each input sample.

  • 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)

  • mod (str, optional) – The cardinality preservation option to use. ("additive", "scaled", "f-additive", "f-scaled", None). (default: None)

  • attention_mechanism (str, optional) – The attention mechanism to use ("within-relation", "across-relation"). (default: "across-relation")

  • attention_mode (str, optional) – The mode to calculate attention logits. ("additive-self-attention", "multiplicative-self-attention"). (default: "additive-self-attention")

  • heads (int, optional) – Number of multi-head-attentions. (default: 1)

  • dim (int) – Number of dimensions for query and key kernels. (default: 1)

  • concat (bool, optional) – If set to False, the multi-head attentions are averaged instead of concatenated. (default: True)

  • negative_slope (float, optional) – LeakyReLU angle of the negative slope. (default: 0.2)

  • dropout (float, optional) – Dropout probability of the normalized attention coefficients which exposes each node to a stochastically sampled neighborhood during training. (default: 0)

  • edge_dim (int, optional) – Edge feature dimensionality (in case there are any). (default: None)

  • 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: Tensor, edge_index: Union[Tensor, SparseTensor], edge_type: Optional[Tensor] = None, edge_attr: Optional[Tensor] = None, size: Optional[Tuple[int, int]] = None, return_attention_weights=None)[source]

Runs the forward pass of the module.

  • x (torch.Tensor) – 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).

  • 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 or torch.sparse.Tensor. (default: None)

  • edge_attr (torch.Tensor, optional) – The edge features. (default: None)

  • size ((int, int), optional) – The shape of the adjacency matrix. (default: None)

  • return_attention_weights (bool, optional) – If set to True, will additionally return the tuple (edge_index, attention_weights), holding the computed attention weights for each edge. (default: None)


Resets all learnable parameters of the module.