sopool-gnns's Introduction

Second-Order Pooling for Graph Neural Networks

This is the code for our paper "Second-Order Pooling for Graph Neural Networks". It is based on the code from GIN. Many thanks!

Created by Zhengyang Wang and Shuiwang Ji at Texas A&M University.

Download & Citation

The paper is now available at IEEE Xplore. If you use our code or results, please kindly cite our paper.

@article{wang2020second,
  author={Wang, Zhengyang and Ji, Shuiwang},
  title={Second-Order Pooling for Graph Neural Networks},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  year={2020},
  publisher={IEEE}
}

System requirement

Programming language

Python 3.6

Python Packages

PyTorch > 1.0.0, tqdm, networkx, numpy

Setup

If you want to try our proposed second-order pooling methods, copy the graphcnn.py file into the models folder from either the sopool_bilinear folder (bilinear second-order pooling) or the sopool_attn folder (attentional second-order pooling).

Run the code

We provide scripts to run the experiments. For bioinformatics datasets and the REDDIT datasets, run

chmod +x run_bio.sh
./run_bio.sh [DATASET] [GPU_ID] [BATCH_SIZE] [HIDDEN_DIM]

For the social network datasets, run

chmod +x run_social.sh
./run_social.sh [DATASET] [GPU_ID] [BATCH_SIZE] [HIDDEN_DIM]

sopool-gnns's People

Contributors

Stargazers

Watchers

sopool-gnns's Issues

Aboue reproducing the results in the paper

Hi，I'm now trying to reproduce the results of your paper.
I tried to adjust hidden_ Dim(16,32,64) and batch_size,(32,128)) .However, in many datasets (such as PTC), there is a certain gap between the my results and this paper. I also try to adjust the random seed, but the results are still not improved.
Do I need to adjust the parameters (lr, layer_num) which hava been given in the paper to achieve the results of the paper?
look forward to your answer

About the experimental settings

Thanks for the code.

I use pytorch-geometric to rewrite your model. I use the same hyperparameters as yours, but I cann't reproduce your method.

In your code, you use the max test acc as the final result. In fact, many existing method use the val set to choose the bese epoch for test set. Many baselines in your paper do like it. So, I think this is a little unfair. Because of this , your results are in the level I can not reach.

zhengyang-wang / sopool-gnns Goto Github PK

sopool-gnns's Introduction

Second-Order Pooling for Graph Neural Networks

Download & Citation

System requirement

Programming language

Python Packages

Setup

Run the code

sopool-gnns's People

Contributors

Stargazers

Watchers

Forkers

sopool-gnns's Issues

Aboue reproducing the results in the paper

About the experimental settings

Any reference paper?

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent