CoNe

This is the PyTorch implementation of Common Neighbor-aware Link Weight Prediction with Simplified Graph Transformer.

Requirements

• Environment:

  • Python == 3.10
  • Nvidia RTX 4090 with CUDA 12.1

• Package dependencies:

  • PyTorch == 2.1.0
  • PyTorch Geometric == 2.4.0

How to run

The commands and hyperparameter settings are provided in the run.sh script. The datasets will be automatically downloaded from the internet during program execution. Make sure the folder data exists in the root directory.

# Neural
python main.py --dataset_name Neural --walk_length 64 --rwpe_use_weight 0 --hidden_channels 64 --num_layers 2 --lr 0.0005 --lr_decay 0.9 --weight_decay 0.1 --batch_size 256

# Celegans
python main.py --dataset_name Celegans --walk_length 64 --rwpe_use_weight 1 --hidden_channels 64 --num_layers 4 --lr 0.005 --lr_decay 0.8 --weight_decay 0.1 --batch_size 64

# Netscience
python main.py --dataset_name Netscience --walk_length 64 --rwpe_use_weight 0 --hidden_channels 64 --num_layers 3 --lr 0.01 --lr_decay 0.95 --weight_decay 0.02 --batch_size 512

# Pblog
python main.py --dataset_name Pblog --walk_length 64 --rwpe_use_weight 0 --hidden_channels 64 --num_layers 3 --lr 0.01 --lr_decay 0.9 --weight_decay 0.05 --batch_size 256

# UCsocial
python main.py --dataset_name UCsocial --walk_length 64 --rwpe_use_weight 0 --hidden_channels 64 --num_layers 3 --lr 0.05 --lr_decay 0.65 --weight_decay 0.2 --batch_size 256

# Condmat
python main.py --dataset_name Condmat --walk_length 16 --rwpe_use_weight 1 --hidden_channels 16 --num_layers 3 --lr 0.001 --lr_decay 0.95 --weight_decay 0.05 --batch_size 512

# Astro
python main.py --dataset_name Astro --walk_length 32 --rwpe_use_weight 0 --hidden_channels 32 --num_layers 2 --lr 0.001 --lr_decay 0.95 --weight_decay 0.05 --batch_size 512

# Collaboration
python main.py --dataset_name Collaboration --walk_length 64 --rwpe_use_weight 0 --hidden_channels 64 --num_layers 3 --lr 0.0001 --lr_decay 0.95 --weight_decay 0.1 --batch_size 256

# Congress
python main.py --dataset_name Congress --walk_length 32 --rwpe_use_weight 0 --hidden_channels 32 --num_layers 3 --lr 0.0005 --lr_decay 0.99 --weight_decay 0.1 --batch_size 1024

# Usair
python main.py --dataset_name Usair --walk_length 128 --rwpe_use_weight 0 --hidden_channels 128 --num_layers 3 --lr 0.001 --lr_decay 0.95 --weight_decay 0.3 --batch_size 64

Baselines

All repositories of the compared methods are listed below.

• Deepwalk & node2vec: https://colab.research.google.com/github/AntonioLonga/PytorchGeometricTutorial/blob/main/Tutorial11/Tutorial11.ipynb

• GraRep & LINE: https://github.com/iamlockelightning/NetworkEmbeddingPytorch

• SDNE: https://github.com/MrLeeeee/SDNE-based-on-Pytorch

• GAE & VGAE: https://github.com/pyg-team/pytorch_geometric/blob/master/examples/autoencoder.py

• GCN: https://github.com/pyg-team/pytorch_geometric/blob/master/examples/gcn.py

• rWCN & rWAA & rWRA: Modified from https://networkx.org/documentation/stable/reference/algorithms/link_prediction.html

• SBM & pWSBM & bWSBM: https://aaronclauset.github.io/wsbm/

• GWNMF: Modified from https://github.com/ahmadvh/Non-Negative-Matrix-factorization---Implemented-in-python/blob/master/The%20notebook.ipynb

• LG-RF & LG-GBDT & LG-SVM: Available upon request from the author via email.

• LWP-WL: https://github.com/unai-zulaika/LWP-WL-Pytorch

• LGLWP: Modified from https://github.com/LeiCaiwsu/LGLP

• Model R: https://github.com/yuchenhou/elephant

• MPLP: Available upon request from the author via email.

• NEW: https://github.com/Lee425/NEW

• SEA: https://github.com/wave-zuo/SEA