GraphCroc

This is the repo of our work, GraphCroc: Cross-Correlation Autoencoder for Graph Structural Reconstruction, in NeurIPS(2024).

🏖️ Environment

GraphCroc is developed on PyG. Generally, please ensure PyTorch >= 1.11.0, python >= 3.8, and pyg >= 2.0.4. Other depedency might be required as well.

🗂️ File Stack

File	Description
/GraphCroc/UNET.py	The two-way Unet model
/GraphCroc/UNET_onebranch.py	The one-way Unet model
/GraphUNET/ops.py	Operations from Graph U-Nets
IMDB_B	A dataset example for GraphCroc (Cross-Correlated)
/IMDB_B/enhance.ipynb	The toy story for different enhancement explored in our work
/IMDB_B/gc_train.py	They downstream task example for graph classification
/IMDB_B/reconstructor.py	The main file to train GraphCroc
/IMDB_B/wl_test_util.py	Utils file for WL-test
/IMDB_B/wl_test.py	WL-test file

🛫 Code Take-off

Current GraphCroc repo is hard coded, and a more structural one is growing.

1. Create folder resLog and tdModels to save results.
2. Main file for training GraphCroc: cd IMDB_B; python3 reconstructor.py
3. WL-test: python3 wl_test.py
4. Graph classification task (100-epoch sample): python3 gc_train.py