Graph_Neural_Networks.md

graph LR
    GAT_Model_Core["GAT Model Core"]
    GATv2_Model_Core["GATv2 Model Core"]
    GAT_Experiment_Management["GAT Experiment Management"]
    GATv2_Experiment_Management["GATv2 Experiment Management"]
    GAT_Experiment_Management -- "instantiates and trains" --> GAT_Model_Core
    GATv2_Experiment_Management -- "instantiates and trains" --> GATv2_Model_Core

Loading

Details

The Graph Neural Networks subsystem focuses on implementing and experimenting with Graph Attention Networks (GAT) and its improved version, GATv2.

GAT Model Core

Implements the core Graph Attention Network (GAT) architecture, defining the neural network layers and attention mechanisms for processing graph-structured data.

Related Classes/Methods:

• labml_nn.graphs.gat.experiment.GAT:108-149

GATv2 Model Core

Implements the core GATv2 architecture, an enhanced version of GAT, providing improved attention mechanisms for graph data.

Related Classes/Methods:

• labml_nn.graphs.gatv2.experiment.GATv2:20-65

GAT Experiment Management

Orchestrates the entire lifecycle of a GAT experiment. This includes loading configurations, handling the Cora dataset, instantiating the GAT model, managing the training and evaluation loops, and logging experiment results.

Related Classes/Methods:

• labml_nn.graphs.gat.experiment.main:288-304
• labml_nn.graphs.gat.experiment.run:194-253
• labml_nn.graphs.gat.experiment.Configs:159-253
• labml_nn.graphs.gat.experiment.CoraDataset:25-105
• labml_nn.graphs.gat.experiment.gat_model:270-275

GATv2 Experiment Management

Manages the complete lifecycle of a GATv2 experiment. This component is responsible for loading configurations, instantiating the GATv2 model, and executing the training and evaluation phases.

Related Classes/Methods:

• labml_nn.graphs.gatv2.experiment.main:90-108
• labml_nn.graphs.gatv2.experiment.run
• labml_nn.graphs.gatv2.experiment.Configs:68-79
• labml_nn.graphs.gatv2.experiment.gat_v2_model:82-87

FAQ