Edge-featured Graph Neural Architecture Search

System Requirements

• Linux
• Python 3.6
• Pytorch 1.9.1
• DGL 0.6.1
• CUDA toolkit 11.3
• One NVIDIA GPU such as RTX 3090.

Run the following command for building the environment.

sudo apt install graphviz
conda env create -f environment.yml
conda activate gnas

Dataset Preparation

Some datasets (CLUSTER, TSP, ZINC, and CIFAR10) are provided by project benchmarking-gnns.

DATASET	TYPE	URL
CLUSTER	node	click here
TSP	edge	click here
ZINC	graph	click here
MNIST	graph	click here
CIFAR10	graph	click here

Search GNN Architectures

We have provided scripts for easily searching graph neural networks on six datasets.

CUDA_VISIBLE_DEVICES=0 python search.py ds=ZINC optimizer=train_optimizer ds.arch_save='archs/TEST' basic.nb_layers=4 basic.nb_nodes=4

Train with Genotypes

We provided scripts for easily training graph neural networks searched by ARGNP.

CUDA_VISIBLE_DEVICES=0 python train.py ds=ZINC  optimizer=train_optimizer ds.load_genotypes='archs/TEST/ZINC/45/cell_geno.txt'