This repository contains an implementation of the Graph Attention Network (GAT) using PyTorch and PyTorch Geometric library. The GAT model is a type of graph neural network that can be used for various graph-related tasks such as node classification, link prediction, and graph classification.
The implementation is based on the original paper: Graph Attention Networks by Petar Velickovic et al.
To install the required dependencies, run:
pip install torch torch-geometric tqdm networkx matplotlib scikit-learn
The main script train.py contains the code for training and testing the GAT model on a given dataset.
After training the model on the cora dataset for 200 epochs, we achieved the following results:
Epoch 200 | Train Loss: 0.022 | Train Acc: 99.17% | Val Loss: 1.50 | Val Acc: 60.80%
Test Accuracy: 79.00%
Training Loss:
Validation Loss:
Training Accuracy:
Validation Accuracy:
Graph after training:
Credits The implementation is based on the following resources:
PyTorch Geometric documentation PyTorch Documentation