Joint Space Learning for Graph Neural Networks (JSGNN) in PyTorch

1. Overview

This repository is an implementation of our paper in PyTorch, based on GIL's implementation. Some of the libraries in the requirements.txt requires manual installation due to older version needed.

2. Setup

2.1 Requirements

python == 3.6.2
torch == 1.1.0
numpy == 1.16.4
scipy == 1.3.0
networkx == 2.3
sage == 9.0
geoopt ==0.0.1
torch_scatter == 1.3.0
torch_geometric == 1.3.0

3. Usage

3.1 set_env.sh

Before training, run

source set_env.sh

This will create environment variables that are used in the code.

3.2 train.py

We provide examples of training commands used to train JSGNN for link prediction and node classification.

Link prediction for GIL

• Cora JSGNN (Test ROC-AUC: 99.36):

python train.py --task lp --dataset cora --model JSGNN --dropout 0.1 --weight-decay 0.0005 --manifold PoincareBall --lr 0.01 --dim 16 --num-layers 1 --act relu --bias 1 --normalize-feats 0

• Pubmed JSGNN (Test ROC-AUC: 96.97):

python train.py --task lp --dataset pubmed --model JSGNN --dropout 0.1 --weight-decay 0.0 --manifold PoincareBall --lr 0.01 --dim 16 --num-layers 1 --act relu --bias 1

Node classification for GIL

• Photo JSGNN (Test accuracy: 97.32):

python train.py --model JSGNN --manifold PoincareBall --lambda-uniform=0.2 --lambda-wasser=0.1 --task nc --dataset amazonphoto --dropout 0.1 --weight-decay 0.0 --lr 0.01 --dim 16 --num-layers 2 --act elu --bias 1 --use-feat=1

• Cora JSGNN (Test accuracy: 83.30):

python train.py --lambda-wasser=0.005 --lambda-uniform=0.5 --task nc --dataset cora --model JSGNN --dropout 0.5 --weight-decay 0.001 --manifold PoincareBall --lr 0.01 --dim 16 --num-layers 3 --act elu --bias 1

Some of the code was forked from the following repositories

• hgcn
• geoopt
• pytorch_geometric
• gil