torch_geometric.nn.conv.GATv2Conv

class GATv2Conv(in_channels: Union[int, Tuple[int, int]], out_channels: int, heads: int = 1, concat: bool = True, negative_slope: float = 0.2, dropout: float = 0.0, add_self_loops: bool = True, edge_dim: Optional[int] = None, fill_value: Union[float, Tensor, str] = 'mean', bias: bool = True, share_weights: bool = False, residual: bool = False, **kwargs)[source]

Bases: MessagePassing

来自“图注意力网络有多专注？”论文的GATv2操作符，它修复了标准GATConv层的静态注意力问题。由于标准GAT中的线性层是紧接着应用的，因此被关注节点的排名与查询节点无关。相比之下，在GATv2中，每个节点都可以关注任何其他节点。

\[\mathbf{x}^{\prime}_i = \sum_{j \in \mathcal{N}(i) \cup \{ i \}} \alpha_{i,j}\mathbf{\Theta}_{t}\mathbf{x}_{j},\]

其中注意力系数 \(\alpha_{i,j}\) 的计算方式为

\[\alpha_{i,j} = \frac{ \exp\left(\mathbf{a}^{\top}\mathrm{LeakyReLU}\left( \mathbf{\Theta}_{s} \mathbf{x}_i + \mathbf{\Theta}_{t} \mathbf{x}_j \right)\right)} {\sum_{k \in \mathcal{N}(i) \cup \{ i \}} \exp\left(\mathbf{a}^{\top}\mathrm{LeakyReLU}\left( \mathbf{\Theta}_{s} \mathbf{x}_i + \mathbf{\Theta}_{t} \mathbf{x}_k \right)\right)}.\]

如果图具有多维边特征 \(\mathbf{e}_{i,j}\)， 则注意力系数 \(\alpha_{i,j}\) 计算如下：

\[\alpha_{i,j} = \frac{ \exp\left(\mathbf{a}^{\top}\mathrm{LeakyReLU}\left( \mathbf{\Theta}_{s} \mathbf{x}_i + \mathbf{\Theta}_{t} \mathbf{x}_j + \mathbf{\Theta}_{e} \mathbf{e}_{i,j} \right)\right)} {\sum_{k \in \mathcal{N}(i) \cup \{ i \}} \exp\left(\mathbf{a}^{\top}\mathrm{LeakyReLU}\left( \mathbf{\Theta}_{s} \mathbf{x}_i + \mathbf{\Theta}_{t} \mathbf{x}_k + \mathbf{\Theta}_{e} \mathbf{e}_{i,k}] \right)\right)}.\]
Parameters:

  • in_channels (int 或 tuple) – 每个输入样本的大小，或 -1 以从 forward 方法的第一个输入推导大小。 在二分图的情况下，元组对应于源维度和目标维度的大小。

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

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

  • add_self_loops (bool, optional) – If set to False, will not add self-loops to the input graph. (default: True)

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

  • fill_value (float 或 torch.Tensor 或 str, 可选) – 生成自环边特征的方式 （在 edge_dim != None 的情况下）。 如果给定为 float 或 torch.Tensor，自环的边特征将直接由 fill_value 给出。 如果给定为 str，自环的边特征将通过聚合指向特定节点的所有边的特征来计算， 根据一个归约操作。（"add", "mean", "min", "max", "mul"）。（默认："mean"）

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

  • share_weights (bool, 可选) – 如果设置为 True，相同的矩阵将应用于每条边的源节点和目标节点，即 \(\mathbf{\Theta}_{s} = \mathbf{\Theta}_{t}\)。 (默认: False)

  • residual (bool, 可选) – 如果设置为 True，该层将添加一个可学习的跳跃连接。（默认值：False）

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

Shapes:

  • input: node features \((|\mathcal{V}|, F_{in})\) or \(((|\mathcal{V_s}|, F_{s}), (|\mathcal{V_t}|, F_{t}))\) if bipartite, edge indices \((2, |\mathcal{E}|)\), edge features \((|\mathcal{E}|, D)\) (optional)

  • 输出: 节点特征 \((|\mathcal{V}|, H * F_{out})\) 或 \(((|\mathcal{V}_t|, H * F_{out})\) 如果是二分图。 如果 return_attention_weights=True, 那么 \(((|\mathcal{V}|, H * F_{out}), ((2, |\mathcal{E}|), (|\mathcal{E}|, H)))\) 或 \(((|\mathcal{V_t}|, H * F_{out}), ((2, |\mathcal{E}|), (|\mathcal{E}|, H)))\) 如果是二分图

forward(x: Union[Tensor, Tuple[Tensor, Tensor]], edge_index: Union[Tensor, SparseTensor], edge_attr: Optional[Tensor] = None, return_attention_weights: Optional[Tensor] = None) → Tensor[source]
forward(x: Union[Tensor, Tuple[Tensor, Tensor]], edge_index: Tensor, edge_attr: Optional[Tensor] = None, return_attention_weights: bool = None) → Tuple[Tensor, Tuple[Tensor, Tensor]]
forward(x: Union[Tensor, Tuple[Tensor, Tensor]], edge_index: SparseTensor, edge_attr: Optional[Tensor] = None, return_attention_weights: bool = None) → Tuple[Tensor, SparseTensor]

运行模块的前向传播。

Parameters:

  • x (torch.Tensor 或 (torch.Tensor, torch.Tensor)) – 输入的节点特征。

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

  • edge_attr (torch.Tensor, optional) – The edge features. (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)

Return type:

Union[Tensor, Tuple[Tensor, Tuple[Tensor, Tensor]], Tuple[Tensor, SparseTensor]]

reset_parameters()[source]

重置模块的所有可学习参数。