GraphCliff: Short-Long Range Gating for Subtle Differences but Critical Changes

This repository contains the reference code for GraphCliff.

Table of Contents

1. Environment Setup
2. Running GraphCliff

Environment Setup

1. Install Miniconda or Anaconda and ensure GPU drivers/CUDA match the versions listed in environment.yml

2. Create the environment:

   cd envs/
   conda env create -f graphcliff.yml
   conda activate graphcliff
   pip install https://data.pyg.org/whl/torch-2.4.0%2Bcu118/torch_scatter-2.1.2%2Bpt24cu118-cp310-cp310-linux_x86_64.whl
   cd ..

Important: Update the prefix at the bottom of the file so it points to your local conda installation before running the command.

Running GraphCliff

Running: Set NUM_GPUS in run.sh to your number of GPUs. Load and validate the best checkpoint if it exists, else train on each data from scratch.

bash run.sh