Graph Attention Networks (GAT) leverage attention mechanisms to learn the importance of neighboring nodes in a graph. This implementation supports both transductive and inductive learning settings.

To install the required dependencies, run:

pip install -r requirements.txt

To train the GAT model on a specific dataset, run:

python GAT/Main.py --dataset <DATASET_NAME>

Replace <DATASET_NAME> with one of the following options: Cora, CiteSeer, Pubmed, PPI.

The supported datasets are:

• Cora: Citation network dataset for node classification.
• CiteSeer: Citation network dataset for node classification.
• Pubmed: Citation network dataset for node classification.
• PPI: Protein-Protein Interaction dataset for multi-class node classification.

The training process varies based on the dataset:

• For transductive learning (e.g., Cora, CiteSeer, Pubmed), the model is trained on the entire graph.
• For inductive learning (e.g., PPI), the model is trained using mini-batch training.

The evaluation metrics include accuracy for transductive learning and F1-score for inductive learning. The evaluation scripts are included in the Train.py file.

The repository includes scripts to visualize the learned node embeddings using UMAP. These visualizations help in understanding the clustering and separation of different node classes after training.

To visualize the embeddings, run the umapplot.py script:

python GAT/umapplot.py

This project is licensed under the MIT License. See the LICENSE file for details.