README.md

GAT (Graph Attention Network)

Introduction

The graph attention network is the follow-up work for GCN. GAT incorporates attention mechanism into GCN. By calculating attention coefficients among nodes, GAT perform a weighted aggregation of node and its neighbors, which allows each node to focus on the most relevant neighbors to make decisions.

Key points to build a GAT

• sample Egographs
• encode EgoGraphs using multi-head GAT convolutional layers.

Original GAT uses full graph as input, for efficient large-scale training, we implemente a sample based version of GAT. For sample based GAT, we use dense format (because sampled number of neighbor nodes are fixed, so they can form a dense tensor) of EgoGraph and for original GAT, we use a sparse format of EgoGraph (we use sparse tensor to deal with unaligned neighbor numbers) is used for batch training.

How to run

1. Prepare data

   cd ../../data/
   python cora.py

2. Train and evaluate

   python train_supervised.py

Datasets and performance

Dataset	ACC
Cora	~0.83

Reference paper

graph attention networks