Demo of Rational Net in spectral domain

func: the target function to approximate
fit: rational function by nerual networks

Rational Neural Network for Graph Signal Prediction

Instruction (under construction)

install required python package

pip install -r requirements.txt  

Choose on pre-processed datasets (Default is 1st below):

• America Revolution (unweighted), pre-processed: TSV file)
• crime (unweighted), pre-processed:TSV file
• language-country (weighted), pre-processed:TSV file

data switch can be configured by pass dataset index in remez_net.py

gen_data(data=1)

Then, run the main program:

python remez_net.py  

GCN configuration is also provided. To perform regression task and compare fairly, neural network weights are removed:

python pygcn/train.py  

MSE

	RationalNet	GCN
America	0.0236	1.3641
crime	0.26021	1.0605
language	0.0329	0.3912

Related papar

Codes for the paper

Zhiqian Chen, Feng Chen, Rongjie Lai, Xuchao Zhang, and Chang-Tien Lu, Rational Neural Networks for Approximating Jump Discontinuities of Graph Convolution Operator, International Conference on Data Mining(ICDM), Singpore, 2018

Citation

@article{chen2018icdmrationalnet
  author    = {Zhiqian Chen and
               Feng Chen and
               Rongjie Lai and
               Xuchao Zhang and
               Chang{-}Tien Lu},
  title     = {Rational Neural Networks for Approximating Jump Discontinuities of Graph Convolution Operator},
  booktitle = {Proceedings of the The IEEE International Conference on Data Mining},
  year      = {2018},
}