README

Code Organization

Training Scripts

citation.py : for three medium-sized datasets, i.e., Cora, Citeseer, and Pubmed

hetero.py: for two heterogeneous datasets, i.e., ACM and DBLP

ogbn.py: for the Ogbn-Products dataset

You can refer to run.sh for specific running commands.

For example, if you want to test the performance of RWGSL with GCN on the Cora dataset, you can run the command as follows:

python citation.py --dataset cora --model GCN --lr 0.1 --hidden_channels 256 --dropout 0.3 --update 5 --walk_len 10 --high 0.95 --low 0.35 --lp_num_layers 2 --lp_alpha 0.4 --first_coe 0.5 --second_coe 0.6 --third_coe 1.4

Particularly, if you need to test the performance of GCN on the original Cora, you can append --train_ori to the above command.

Functional Codes

sample.py: neighborhood sampling, similarity calculation, random walk, and structure modification

metrics.py: accuracy and $F_1$ score calculation

models.py: the definitions of neural network classes

normalization.py: different kinds of matrix normalization functions

utils.py: utility functions for data preparation, model instantiation, etc.

Data

All data should be placed into the ./data directory.