benchmarks for gcns
if you want to use pre_processed bpgcn version on pubmed you can run:
python bpgcn_real.py --epoch 200 --early_stop 25 --eps1 0.1 --eps2 0.6 --cuda 2 --netdepth 12 --beta 0.5 --dataset pubmed --vanilla 0 --bib 0
Also you can get classification of cora and citeseer using :
python bpgcn_real.py --epoch 200 --early_stop 25 --eps1 0.1 --eps2 0.6 --cuda 2 --netdepth 5 --beta 2 --dataset cora
python bpgcn_real.py --epoch 200 --early_stop 25 --eps1 0.1 --eps2 0.5 --cuda 2 --netdepth 5 --beta 2 --dataset citeseer
if you want to get Fig.4 you can run :
chmod 777 run_syn.sh(maybe not needed)
./run_syn.sh
BP with C++ can be run in the file JSBM_C++ with command:
./test1.sh
for Fig3, may using MPI for multipul instances on varying parameters.