Graph Wavelet Neural Network. Bingbing Xu, Huawei Shen, Qi Cao, Yunqi Qiu, Xueqi Cheng. ICLR, 2019. pdf
We provide a TensorFlow implementation of Graph Wavelet Neural Network, which implements graph convolution via graph wavelet transform instead of Fourier transform. Different from graph Fourier transform, graph wavelets are sparse and localized in vertex domain, offering high efficiency and good interpretability for graph convolution. We evaluated our model in the task of graph-based semi-supervised classification.
the script has been tested running under Python 2.7, with the following packages installed (along with their dependencies):
- tensorflow==0.12.0
- numpy==1.14.0
- scipy==0.19.1
- networkx==2.0
- cd GraphWaveletNetwork
- python train.py
- --wavelet_s FLOAT wavelet scaling parameter. Default: Cora: 1.0, Citeseer: 0.7, Pubmed: 0.5
- --threshold FLOAT threshold parameter for wavelet. Default: Cora: 1e-4, Citeseer: 1e-5, Pubmed: 1e-7
- --epochs INT Number of Adam epochs. Default: 1000.
- --early-stopping INT Number of early stopping epochs. Default: 100.
The run example for Cora dataset in default parameter
Please cite our paper if you use this code in your own work:
@inproceedings{ xu2018graph, title={Graph Wavelet Neural Network}, author={Bingbing Xu and Huawei Shen and Qi Cao and Yunqi Qiu and Xueqi Cheng}, booktitle={International Conference on Learning Representations}, year={2019}, url={https://openreview.net/forum?id=H1ewdiR5tQ}, }
Some sections of code adapted from tkipf/gcn(https://github.com/tkipf/gcn)