/alpha-SPE

Codebase of paper "Balancing structure and position information in Graph Transformer network with a learnable node embedding"

Primary LanguagePython

Implementation of Generalized Structural and Positional Encoding

learn a joint node encoding that preserved both local node structural and global graph structure in Graph Transformer

Install

  • The code is run with the dgl library (https://docs.dgl.ai/).
  • The code base on the benchmarking graph neural network repo (https://github.com/graphdeeplearning/benchmarking-gnns.git).
  • Due to the legacy code of dgl, several code base is run on dgl==0.9.1 and others run on dgl=1.0.0 with or without GPU supports
  • Support baseline:
    • Standard GNN models (GCN, GAT):
    • Graph Transformer based models: