#

Train a Graph Attention Network v2 (GATv2) on Cora dataset

13import torch
14from torch import nn
15
16from labml import experiment
17from labml.configs import option
18from labml_helpers.module import Module
19from labml_nn.graphs.gat.experiment import Configs as GATConfigs
20from labml_nn.graphs.gatv2 import GraphAttentionV2Layer

#

Graph Attention Network v2 (GATv2)

This graph attention network has two graph attention layers.

23class GATv2(Module):

#

in_features is the number of features per node
n_hidden is the number of features in the first graph attention layer
n_classes is the number of classes
n_heads is the number of heads in the graph attention layers
dropout is the dropout probability
share_weights if set to True, the same matrix will be applied to the source and the target node of every edge

30    def __init__(self, in_features: int, n_hidden: int, n_classes: int, n_heads: int, dropout: float,
31                 share_weights: bool = True):

#

40        super().__init__()

#

First graph attention layer where we concatenate the heads

43        self.layer1 = GraphAttentionV2Layer(in_features, n_hidden, n_heads,
44                                            is_concat=True, dropout=dropout, share_weights=share_weights)

#

Activation function after first graph attention layer

46        self.activation = nn.ELU()

#

Final graph attention layer where we average the heads

48        self.output = GraphAttentionV2Layer(n_hidden, n_classes, 1,
49                                            is_concat=False, dropout=dropout, share_weights=share_weights)

#

Dropout

51        self.dropout = nn.Dropout(dropout)

#

x is the features vectors of shape [n_nodes, in_features]
adj_mat is the adjacency matrix of the form [n_nodes, n_nodes, n_heads] or [n_nodes, n_nodes, 1]

53    def forward(self, x: torch.Tensor, adj_mat: torch.Tensor):

#

Apply dropout to the input

60        x = self.dropout(x)

#

First graph attention layer

62        x = self.layer1(x, adj_mat)

#

Activation function

64        x = self.activation(x)

#

Dropout

66        x = self.dropout(x)

#

Output layer (without activation) for logits

68        return self.output(x, adj_mat)

#

Configurations

Since the experiment is same as GAT experiment but with GATv2 model we extend the same configs and change the model.

71class Configs(GATConfigs):

#

Whether to share weights for source and target nodes of edges

80    share_weights: bool = False

#

Set the model

82    model: GATv2 = 'gat_v2_model'

#

Create GATv2 model

85@option(Configs.model)
86def gat_v2_model(c: Configs):

#

90    return GATv2(c.in_features, c.n_hidden, c.n_classes, c.n_heads, c.dropout, c.share_weights).to(c.device)

#

93def main():

#

Create configurations

95    conf = Configs()

#

Create an experiment

97    experiment.create(name='gatv2')

#

Calculate configurations.

99    experiment.configs(conf, {

#

Adam optimizer

101        'optimizer.optimizer': 'Adam',
102        'optimizer.learning_rate': 5e-3,
103        'optimizer.weight_decay': 5e-4,
104
105        'dropout': 0.7,
106    })

#

Start and watch the experiment

109    with experiment.start():

#

Run the training

111        conf.run()

#

115if __name__ == '__main__':
116    main()