EdgeGatedGraphConv

class nfflr.nn.EdgeGatedGraphConv(input_features: int, output_features: int, residual: bool = True, norm: Literal['layernorm', 'batchnorm', 'instancenorm'] = 'layernorm', skip_edgenorm: bool = False)

¹Xavier Bresson and Thomas Laurent. Residual gated graph convnets. 2018. arXiv:1711.07553.²Vijay Prakash Dwivedi, Chaitanya K. Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking graph neural networks. Journal of Machine Learning Research, 24(43):1–48, 2023. URL: http://jmlr.org/papers/v24/22-0567.html.³Tian Xie and Jeffrey C. Grossman. Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties. Physical Review Letters, April 2018. doi:10.1103/physrevlett.120.145301.

Edge gated graph convolution

See Bresson and Laurent ^{1 1}Xavier Bresson and Thomas Laurent. Residual gated graph convnets. 2018. arXiv:1711.07553. for reference, and refer to Dwivedi et al. ^{2 2}Vijay Prakash Dwivedi, Chaitanya K. Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking graph neural networks. Journal of Machine Learning Research, 24(43):1–48, 2023. URL: http://jmlr.org/papers/v24/22-0567.html. for detailed discussion.

\[x_i^{l+1} = SiLU ( U x_i^l + \sum_{j \in \mathcal{N}(i)} \eta_{ij} ⊙ V x_j^l)\]

This is similar to the interaction from CGCNN ^{3 3}Tian Xie and Jeffrey C. Grossman. Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties. Physical Review Letters, April 2018. doi:10.1103/physrevlett.120.145301., but edge features only go into the soft attention / edge gating function, and the primary node update function is W cat(u, v) + b

Parameters:

input_featuresint

output_featuresint

residualbool, default=True

add skip connection for both node and edge features

norm{“layernorm”, “batchnorm”, “instancenorm”}, optional

skip_edgenormbool default=False

omit normalization of edge features

Methods

forward(g, node_feats, edge_feats)

Edge-gated graph convolution.

forward(g: DGLGraph, node_feats: Tensor, edge_feats: Tensor) → tuple[Tensor, Tensor]

Edge-gated graph convolution.

Parameters:

gdgl.DGLGraph

input graph

node_featstorch.Tensor

input node features

edge_featstorch.Tensor

input edge features

Returns:

node_featurestorch.Tensor

edge_featurestorch.Tensor