DEAL

Overview

This is the implementation of "Inductive Link Prediction for Nodes Having Only Attribute Information " published at IJCAI 2020.

This work has been adapted from the original implementation by working-yuhao, found at https://github.com/working-yuhao/DEAL.

In particular, this work is a refactoring of the original implementation to be more minimal and composable. This refactoring work focuses solely on the inductive link prediction task.

Major changes

Datasets are not the same as in the original paper
- Graphs are undirected (twice the number of edges)
- Different train/validation/test splits
- Different method for negative sampling, negative sampling fraction

Requirements

Linux
Nvidia GPU
Cuda version 11.0
Anaconda

Installation

The installation below uses anaconda.

#!/bin/bash

CONDA_ENV_NAME=DEAL

#conda clean --all -y
conda env list | grep ${CONDA_ENV_NAME}
if [ $? -eq 0 ]; then
    echo "DEAL environment already exists; skipping creation"
else
    echo "Creating ${CONDA_ENV_NAME} environment"
    conda create -n ${CONDA_ENV_NAME} python=3.9 -y
fi

conda activate ${CONDA_ENV_NAME}

conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch -c nvidia -y
conda install -q -y numpy pyyaml scipy ipython mkl mkl-include conda-build
conda install pyg -c pyg -c conda-forge -y

pip install -e .

Original Paper DEAL Inductive Results

** Note: The below table is not a fair comparison (due to the changes to the dataset mentioned above)

Implementation	Cora	Cora-full	CiteSeer	CS	PubMed	Computers	Photos
	AUC/AP	AUC/AP	AUC/AP	AUC/AP	AUC/AP	AUC/AP	AUC/AP
Original	0.864/0.804	N/A	0.937/0.907	0.977/0.959	0.966/0.931	0.953/0.899	0.965/0.922
Refactored	0.8757/0.8753	0.9124/0.9095	0.9114/0.9160	0.9485/0.9412	0.9226/0.9140	0.8494/0.8373	0.8920/0.8804

Reproduction

Cora (small)

make train-cora-small

 Total Load data time: 0.02 s
 Total Train/val time: 6.39 s
 Test time: 0.00 s
 Total time: 6.40 s
 ROC-AUC: 0.8757 
 AP: 0.8753

Cora (full)

make train-cora-full

 Total Load data time: 0.57 s
 Total Train/val time: 33.38 s
 Test time: 0.00 s
 Total time: 33.95 s
 ROC-AUC: 0.9124
 AP: 0.9095

CiteSeer

make train-citeseer

 Total Load data time: 0.04 s
 Total Train/val time: 9.83 s
 Test time: 0.00 s
 Total time: 9.87 s
 ROC-AUC: 0.9114 
 AP: 0.9160

Co-authorship (CS)

make train-coauthor-cs

 Total Load data time: 0.48 s
 Total Train/val time: 41.70 s
 Test time: 0.00 s
 Total time: 42.18 s
 ROC-AUC: 0.9485 
 AP: 0.9412

Pubmed

make train-pubmed

 Total Load data time: 120.92 s
 Total Train/val time: 20.63 s
 Test time: 0.00 s
 Total time: 141.55 s
 ROC-AUC: 0.9226
 AP: 0.9140

Amazon Computers

make train-computers

 Total Load data time: 0.28 s
 Total Train/val time: 69.72 s
 Test time: 0.00 s
 Total time: 70.00 s
 ROC-AUC: 0.8494 
 AP: 0.8373

Amazon Photos

make train-photos

 Total Load data time: 0.12 s
 Total Train/val time: 30.83 s
 Test time: 0.00 s
 Total time: 30.95 s
 ROC-AUC: 0.8920
 AP: 0.8804

Citing the original work

Please cite the original IJCAI 2020 paper:

@inproceedings{ijcai2020-168,
  title     = {Inductive Link Prediction for Nodes Having Only Attribute Information},
  author    = {Hao, Yu and Cao, Xin and Fang, Yixiang and Xie, Xike and Wang, Sibo},
  booktitle = {Proceedings of the Twenty-Ninth International Joint Conference on
               Artificial Intelligence, {IJCAI-20}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},             
  editor    = {Christian Bessiere},	
  pages     = {1209--1215},
  year      = {2020},
  month     = {7},
  note      = {Main track}
  doi       = {10.24963/ijcai.2020/168},
  url       = {https://doi.org/10.24963/ijcai.2020/168},
}

lajd/DEAL