This is a PyTorch implementation of the paper "SPAGAN: Shortest Path Graph Attention Network"

We prefer to create a new conda environment to run the code.

Version >= 1.0.0

We use torch_geometric, torch_scatter and torch_sparse as backbone to implement the path attention mechanism. Please follow the official website to install them.

We use networkx to load the graph dataset.

We use graph-tool for fast APSP calculation. Please follow the official website to install.

Activating the corresponding conda env if you use a new conda environment.

python train_spgat.py