Source code for pygip.models.nn.backbones

import torch
import torch.nn as nn
import torch.nn.functional as F
from dgl.nn.pytorch import GraphConv, SAGEConv
from torch_geometric.nn import GATConv
from torch_geometric.nn import GCNConv


[docs]class GCN(nn.Module): """A simple GCN Network.""" def __init__(self, feature_number, label_number): super(GCN, self).__init__() self.layers = nn.ModuleList() self.layers.append(GraphConv(feature_number, 16, activation=F.relu)) self.layers.append(GraphConv(16, label_number)) self.dropout = nn.Dropout(p=0.5)
[docs] def forward(self, g, features): x = self.layers[0](g, features) x = F.relu(x) x = self.layers[1](g, x) return x
[docs]class GraphSAGE(nn.Module): """ A GraphSAGE model implemented with PyG's SAGEConv module. It consists of two SAGEConv layers: - The first layer projects features to 'hidden_channels', - The second layer outputs 'out_channels'. """ def __init__(self, in_channels, hidden_channels, out_channels): """ Initializes the GraphSAGE model. Parameters ---------- in_channels : int The dimensionality of the input features. hidden_channels : int The dimensionality of the hidden layer. out_channels : int The dimensionality of the output layer (or the number of classes). """ super(GraphSAGE, self).__init__() self.conv1 = SAGEConv(in_channels, hidden_channels, aggregator_type='mean') self.conv2 = SAGEConv(hidden_channels, out_channels, aggregator_type='mean')
[docs] def forward(self, blocks, x): """ Forward pass. Parameters ---------- blocks : list of dgl.DGLGraph A list of subgraphs sampled for multiple layers. x : torch.Tensor The node features of shape (num_nodes, in_channels). Returns ------- torch.Tensor The model outputs (logits) of shape (num_nodes, out_channels). """ x = self.conv1(blocks[0], x) x = F.relu(x) x = self.conv2(blocks[1], x) return x
[docs]class ShadowNet(torch.nn.Module): """A shadow model GCN.""" def __init__(self, feature_number, label_number): super(ShadowNet, self).__init__() self.layer1 = GraphConv(feature_number, 16) self.layer2 = GraphConv(16, label_number)
[docs] def forward(self, g, features): x = torch.nn.functional.relu(self.layer1(g, features)) x = self.layer2(g, x) return x
[docs]class AttackNet(nn.Module): """An attack model GCN.""" def __init__(self, feature_number, label_number): super(AttackNet, self).__init__() self.layers = nn.ModuleList() self.layers.append(GraphConv(feature_number, 16, activation=F.relu)) self.layers.append(GraphConv(16, label_number)) self.dropout = nn.Dropout(p=0.5)
[docs] def forward(self, g, features): x = F.relu(self.layers[0](g, features)) x = self.layers[1](g, x) return x
[docs]class GAT(nn.Module): def __init__(self, in_channels, hidden_channels, out_channels, heads=8): super().__init__() self.conv1 = GATConv(in_channels, hidden_channels, heads=heads) self.conv2 = GATConv(hidden_channels*heads, out_channels, heads=1)
[docs] def forward(self, x, edge_index): x = F.relu(self.conv1(x, edge_index)) return self.conv2(x, edge_index)
[docs]class GCN_PyG(nn.Module): # Rename to avoid clash with existing DGL GCN def __init__(self, in_channels, hidden_channels, out_channels): super().__init__() self.conv1 = GCNConv(in_channels, hidden_channels) self.conv2 = GCNConv(hidden_channels, out_channels)
[docs] def forward(self, x, edge_index): x = F.relu(self.conv1(x, edge_index)) return self.conv2(x, edge_index)