Local Augmentation for Graph Neural Networks

This repository contains experiments done as part of course project for the paper "Local Augmentation for Graph Neural Networks".

Dependencies

• CUDA 10.2.89
• python 3.6.8
• pytorch 1.9.0
• pyg 2.0.3

Usage

• For semi-supervised setting, run the following script

cd Citation
bash semi.sh

• For full-supervised setting, run the following script

cd OGB
# If you want to pre-train the generative model, run the following command:
python cvae_generate_products.py --latent_size 10 --pretrain_lr 1e-5 --total_iterations 10000 --batch_size 8192
# Train downstream GNNs
bash full.sh

Additions :

1. Graph Classification : @Aziz-Shameem

CVAE Pretrining : Run cvae_train.py with the appropriate parameters
Model Training : Run lagin_graphlevel.py with the appropriate parameters (after pretraining the CVAE)

cd Citation
# for pretraining the CVAE
python cvae_train.py
# for training the GIN model
python lagin_graphlevel.py

2. Link Prediction: @BhavyaKohli

CVAE Pretraining : (required for LINKXDatasets, optional for retraining Planetoid datasets) run cvae_train_linkx.py or cvae_generate_citation.py with required parameters, for LINKX and Planetoid datasets respectively

Model Training : Run lagcn_linkpred.py with the appropriate parameters (after pretraining the CVAE)

cd Citation
# for pretraining the CVAE
python cvae_train_linkx.py --dataset "reed98"
# for training the GIN model
python lagcn_linkpred.py

3. Normalizing-Flow Model: @prateekgargx

For semi-supervised setting, we provide two pre-trained generative models: Conditional VAE, and Conditional Normalizing Flow. If you want to pre-train yourself, use:

cd Citation
python cvae_generate_citation.py --model 1 [other-parameters]
# 0 for CVAE, 1 for CNF