Skip to contents

Multi-layer GCN Model

model_gcn.Rd

Stacks multiple GCN layers to create a deep graph convolutional network.

Usage

model_gcn(
  in_features,
  hidden_dims,
  out_features,
  activation = nnf_relu,
  out_activation = NULL,
  dropout = 0,
  normalize = TRUE
)

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)

activation

Function. Activation for hidden layers. Default: nnf_relu

dropout

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

normalize

Logical. Whether to add self-loops and apply symmetric normalization. Default: TRUE

output_activation

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

x

Tensor n_nodes x in_features. Node feature matrix

adj

Tensor n_nodes x n_nodes. Binary adjacency matrix (0/1) defining graph structure. When normalize = TRUE, self-loops are added and symmetric normalization is applied automatically

edge_weight

Tensor n_nodes x n_nodes or NULL. Optional edge weights to apply to the adjacency structure. If NULL, treats all edges as having weight 1. Passed through to all layers. Default: NULL

Value

Tensor n_nodes x out_features. Final predictions

Details

Architecture:

  • L hidden GCN layers with configurable activation

  • 1 output GCN layer with optional output activation

  • Total layers = length(hidden_dims) + 1

For example, hidden_dims = c(56, 56) creates:

  • Layer 1: in_features → 56 (activation)

  • Layer 2: 56 → 56 (activation)

  • Layer 3: 56 → out_features (output_activation)

Uses gcn_conv_layer which automatically handles adding self-loops and symmetric normalization when normalize = TRUE.

Forward pass

Examples

if (FALSE) { # \dontrun{
# Binary classification with sigmoid
model <- model_gcn(14, c(56, 56), 1, output_activation = nnf_sigmoid)

# Multi-class with softmax
model <- model_gcn(
  14,
  c(32, 32),
  10,
  output_activation = function(x) nnf_softmax(x, dim = -1)
)

# Regression (no output activation)
model <- model_gcn(14, c(64, 64), 1)

# With dropout and tanh activation
model <- model_gcn(14, c(56, 56), 1,
                   activation = nnf_tanh,
                   dropout = 0.5)
} # }