carrtesy/GNN

This repo implements graph neural network architectures.

GNN

This repo reimplements graph neural network architectures.

Quickstart

python main.py --dataset {dataset_name}

for example,

python main.py --dataset cora

Dataset

For now, most of the dataset pipeline is from: https://github.com/tkipf/gcn

• citeseer [explaination] [rawdata]
• cora [explaination] [rawdata]
• pubmed [explaination] [rawdata]
• NELL (To be updated)

Architectures

• GCN: Kipf, T. N., & Welling, M. (2017). Semi-Supervised Classification with Graph Convolutional Networks. International Conference on Learning Representations (ICLR).

Experiments

Numbers in parenthesis states results from original paper.

Method	Citeseer	Cora	Pubmed	NELL
GCN [Kipf & Welling]	68.9 (70.3)	80.79 (81.5)	78.9 (79.0)

Details

GCN(Kipf)

• (Citeseer) 100 epochs, dropout rate = 0.5, L2 regularization with 5e-04, num_hiddens of 16
• (Cora) 100 epochs, dropout rate = 0.5, L2 regularization with 5e-04, num_hiddens of 16
• (Pubmed) 100 epochs, dropout rate = 0.5, L2 regularization with 5e-04, num_hiddens of 16