AutoCoG: A Unified Data-Model Co-Search Framework for Graph Neural Networks

Code used for AutoCoG: A Unified Data-Model Co-search Framework for Graph Neural Networks

Installation

We recommend users to use conda to install the running environment. Use the following cmd to install the enviroment:

conda create --file env.yml

Usage

After installing the enviroment, users can directly train the neural networks by invoking:

bash scripts/train_script.sh

Immediately they will be prompt the following:

• cuda (please supply the numerical id value of your gpu)
• dataset (refers to ./configurations for supported dataset, supply the dataset name in lowercase here)
• outputdir (supply the output path)
• n_layers (supply the desired number of layers for GNN)
• p_stages (supply the desired number of progressive search stages).

For example:

cuda: 0
dataset: texas
outputdir: output/
n_layers: 8
p_stages: 4

Citation

if you find this repo is helpful, please cite

@inproceedings{
hoang2022autocog,
title={AutoCoG: A Unified Data-Model Co-Search Framework for Graph Neural Networks},
author={Duc N.M Hoang and Kaixiong Zhou and Tianlong Chen and Xia Hu and Zhangyang Wang},
booktitle={First Conference on Automated Machine Learning (Main Track)},
year={2022},
url={https://openreview.net/forum?id=r0zIWWar8gq}
}