GNN-Asymptotically-Constant

This repository contains the official code base of the paper Almost Surely Asymptotically Constant Graph Neural Networks.

Installation

To reproduce the results please use Python 3.9, PyTorch version 2.0.0, Cuda 11.8, PyG version 2.3.0, and torchmetrics.

pip install torch==2.0.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.0.0+cu118.html
pip install torch-geometric==2.3.0
pip install ogb

Datasets

datasets

ER, LogER and InverseER represent the $ER(n, p(n) = 0.1)$, $ER(n, p(n) = \frac{\log{n}}{n})$ and $ER(n, p(n) = \frac{1}{50n})$ distributions.

Running

The script we use to run the experiments is ./main.py. Note that the script should be run with . as the main directory or source root.

The parameters of the script are:

• --dataset: name of the dataset. The available options are: ER, LogER, InverseER, SBM, BA, Tiger1k, Tiger5k, Tiger10k, Tiger25k and Tiger90k.

• --graph_size: the graph size.

• --num_graph_samples: the number of different graph size samples taken.

• --rw_pos_length: the maximal length of the random walk in the Random Walk Positional Encoding.

• --model_type: the type of model that is used. The available options are: MEAN_GNN, GCN, GAT and GPS.

• --num_layers: the network's number of layers.

• --in_dim: the network's input dimension.

• --hidden_dim: the network's hidden dimension.

• --output_dim: the network's output dimension.

• --pool: name of the graph pooling.

• --seed: a seed to set random processes.

• --gpu: the number of the gpu that is used to run the code on.

Example running

To perform experiments over the LogER dataset with a MEAN_GNN with 3 layers, output dimension of 5 and an input and hidden dimension of 128. See an example for the use of the following command:

python -u main.py --dataset LogER --model_type MEAN_GNN --in_dim 128 --hidden_dim 128 --out_dim 5 --num_layers 3 --seed 0