This is the implementation of paper:

Ripple Walk Training: A Subgraph-based training framework for Large and Deep Graph Neural Network

Requirements

The code is implemented in Python 3.7. Package used for development are just below.

networkx           
numpy              
scipy              
torch              

Datasets

Pubmed

###Instructions for running the code

1, Run the subgraph sampling code

python3 subgraph_sample_pubmed.py

, the results will be stored in ./sampled_subgraph/.

2, Run the GCN or GAT model training/testing code

python3 train_rw_gcn_pubmed.py

or

python3 train_rw_gat_pubmed.py

, the results will be shown on screen and stored in ./results/.

###Note:

1, If no GPU is available, add config --no-cuda True when running the GCN/GAT models. 2, To change epoch numbers (default as 10) of training to NUMBER, add config --epochs NUMBER.