GraphRNN

Codes for the paper "Learning Graph-Level Representations with Gated Recurrent Neural Networks". (https://arxiv.org/pdf/1805.07683.pdf)

1. Setup

Get the source code,

git clone https://github.com/yuj-umd/graphRNN.git

Install pytorch from https://pytorch.org/

2. Usage

Run

python main.py \
        -seed 1 \
    -data $data \
    -learning_rate $learning_rate \
    -num_epochs $num_epochs \
    -hidden $hidden \
    -fold $fold \
    -embedding_dim $embedding_dim \
    -rnn_hidden_dim $rnn_hidden_dim 

Paramaters are defined as

data: MUTAG, NCI1, NCI109, DD, ENZYMES
feat_dim: Number of node labels
embedding_dim: Dimension of node embedding
num_class: Number of graph classes
rnn_hidden_dim: Hidden unit size of RNN
learning_rate: initial learning_rate

3. Reference

@article{jin2018learning,
    title={Learning Graph-Level Representations with Gated Recurrent Neural Networks},
    author={Jin, Yu and JaJa, Joseph F},
    journal={arXiv preprint arXiv:1805.07683},
    year={2018}
 }