Multi-layer Graph Isomorphism Network Model (Xu et al. 2019)

Stacks multiple GIN layers with MLP transformations.

Usage

model_gin(
  in_features,
  hidden_dims,
  out_features,
  eps = 0,
  learn_eps = FALSE,
  dropout = 0.5,
  out_activation = NULL
)

Arguments

in_features

Integer. Number of input features per node

hidden_dims

Integer vector. Dimensions of hidden layers (length = L)

out_features

Integer. Number of output features (typically 1 for regression)

eps

Numeric. Initial value for epsilon. Default: 0

learn_eps

Logical. Whether to learn epsilon parameters. Default: FALSE

dropout

Numeric. Dropout rate (0-1) applied between layers. Default: 0.5

out_activation

Function or NULL. Activation for output layer. Default: NULL

Value

Tensor n_nodes x out_features. Final predictions

Details

Architecture:

• L hidden GIN layers with MLP transformations

• 1 output GIN layer with optional output activation

• Total layers = length(hidden_dims) + 1

Each layer aggregates neighbor features via summation, adds weighted self features, and applies a 2-layer MLP transformation.

Forward pass

model(x, adj)

• x: Tensor n_nodes x in_features. Node feature matrix.

• adj: Sparse COO tensor n_nodes x n_nodes. Adjacency matrix defining graph structure.

References

Xu, K., Hu, W., Leskovec, J., & Jegelka, S. (2019). How Powerful are Graph Neural Networks? International Conference on Learning Representations. doi:10.48550/arXiv.1810.00826

Examples

if (FALSE) { # torch::torch_is_installed()
adj <- adj_from_edgelist(from = c(1, 2, 3, 4), to = c(2, 3, 4, 1))
x <- torch::torch_randn(4, 14)

# Binary classification
model <- model_gin(14, c(64, 64), 1, out_activation = torch::nnf_sigmoid)
model(x, adj)

# Multi-class classification
model <- model_gin(
  14,
  c(64, 64),
  3,
  out_activation = function(x) torch::nnf_softmax(x, dim = -1)
)
model(x, adj)

# With learnable epsilon
model_gin(14, c(128), 1, learn_eps = TRUE)
}