deepGCFX

PyTorch implementation for our AAAI 2022 Paper "Graph-wise Common Latent Factor Extraction for Unsupervised Graph Representation Learning"

Preparing the environment

1. Our implementation is in PyTorch running on GPUs. Use the provided environment.yml to create a virtual environment using Anaconda.
2. Commands to automatically download graph-level and node-level datasets from Pytorch-Geometric are included in the main.py in each folder.

Implementation Details

This repository contains implementations for all experiments we have used in our paper. deepgcfx_graph folder contains the implimentation for graph level tasks while deepgcfx_node contains node-level implimentation.

Final hyper-parameter values to reproduce our results are provided in the Supplementary of our paper and will be updated in this repository soon.

Training Steps

Go inside each respective folder and execute following commands.

Graph-level

python -u main.py --DS dataset_name --lr 0.001 --num-gc-layers 3 --hidden-dim 128 --batch_size 128 --num_epochs 500 --model_name deepgcfx_graph

Node-level

python -u main.py --lr 0.001 --num_epochs 2000 --num-gc-layers 1 --hidden-dim 512 --model_name deepgcfx_node