Skip to contents

Multi-layer GraphSAGE Model (Hamilton et al. 2017)

Source: R/model-sage.R
model_sage.Rd

Stacks multiple GraphSAGE layers with configurable aggregation functions.

Usage

model_sage(
  in_features,
  hidden_dims,
  out_features,
  aggregator = MeanAggregator(),
  activation = nnf_relu,
  out_activation = NULL,
  dropout = 0,
  concat = TRUE,
  norm = 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)

aggregator

Aggregator S7 object. Aggregation function for all layers. Default: MeanAggregator()

activation

Function. Activation for hidden layers. Default: nnf_relu

out_activation

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

dropout

Numeric. Dropout rate (0-1) applied after each hidden layer. Default: 0

concat

Logical. If TRUE, concatenates self and neighbor features. If FALSE, adds them. Default: TRUE

norm

nn_module generator or NULL. Normalization applied after each hidden layer, before the activation. Called once per hidden layer with that layer's output dimension. Default: NULL

Value

Tensor n_nodes x out_features. Final predictions

Details

Architecture:

  • L hidden SAGE layers with configurable activation

  • 1 output SAGE layer with optional output activation

  • Total layers = length(hidden_dims) + 1

Each layer aggregates neighbor features using the specified aggregator, then combines with self features via concatenation or addition.

Forward pass

model(x, adj, batch = NULL)

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

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

  • batch: Tensor or NULL. Batch vector assigning each node to a graph, using 1-based graph indices. Passed to norm. If NULL, all nodes are treated as a single graph.

References

Hamilton, W., Ying, Z., & Leskovec, J. (2017). Inductive representation learning on large graphs. Advances in Neural Information Processing Systems, 30. doi:10.48550/arXiv.1706.02216

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 with sigmoid and mean aggregation
model <- model_sage(14, c(56, 56), 1, out_activation = torch::nnf_sigmoid)
model(x, adj)

# Multi-class with softmax and max aggregation
model <- model_sage(
  14,
  c(32, 32),
  10,
  aggregator = MaxAggregator(),
  out_activation = function(x) torch::nnf_softmax(x, dim = -1)
)
model(x, adj)

# Regression with sum aggregation
model_sage(14, c(64, 64), 1, aggregator = SumAggregator())

# With dropout and custom activation
model_sage(14, c(56, 56), 1, activation = torch::torch_tanh, dropout = 0.5)

# With normalization after each hidden layer
model <- model_sage(14, c(56, 32), 1, norm = layer_layer_norm)
model(x, adj)

model_sage(14, c(56, 32), 1, norm = \(d) layer_layer_norm(d, mode = "node"))
}