README.md

GCN

Graph Convolution Network for Pytorch

GCN implementation for paper: Semi-Supervised Classification with Graph Convolutional Networks

Referenced from these sources

• original pygcn only available for Cora [link]
• data processing for other datasets [link]
• preprocessing NELL dataset [link]

Benchmark

Task: Semi-supervised & Inductive learning

Dataset	This repository (default seed 42)
citeseer	0.7030
cora	0.8020
pubmed	0.7940
nell	0.58

Configurations for Nell

'dataset'       : 'nell.0.001',
'model'         : 'gcn',
'learning_rate' : 0.01,
'epochs'        : 200,
''hidden1'      : 64,
'dropout'       : 0.1,
'weight_decay'  : 1e-5,

Configurations for other datasets

'dataset'       : '',
'model'         : 'gcn',
'learning_rate' : 0.01,
'epochs'        : 200,
''hidden1'      : 16,
'dropout'       : 0.5,
'weight_decay'  : 5e-4,

How to start

pip install .
python3 train.py --dataset=nell.0.001 --hidden=64 --weight_decay=1e-5 --dropout=0.1
python3 train.py --dataset=cora
python3 train.py --dataset=pubmed
python3 train.py --dataset=citeseer