LitGNN: Unleash the Power of Graph Neural Networks

Experiment, Optimize, and Innovate with Ease.

---

Welcome to LitGNN

LitGNN is a GNN experimentation playground that combines the best of:

• 🐉Hydra: Effortless CLI and configuration management.
• 📊Weights and Biases: Track experiments and artifacts with ease.
• 🎯Optuna: Hyperparameter optimization made simple with seamless Hydra integration.
• 🕸️PyTorch Geometric (PyG): Implement and explore state-of-the-art GNN models and datasets.
• ⚡PyTorch Lightning: Streamline your workflow with lightning-fast training.

Following datasets are currently supported,

Dataset type	Group	Datasets
molecule_net		PyG MoleculeNet datasets
custom	TDC	ADMET Benchmark datasets
custom	Biogen	ADME datasets

Get started

Create a .env file similar to .env.example file, and fill out the values for each variables in the .env file.

🚀 Installation

Note

The following shell script will create a conda env named litgnn and install the necessary dependencies.

# CPU (default)
bash scripts/setup_env.sh
# GPU
bash scripts/setup_env.sh cu118

💻 Training recipes

Browse all training recipes here.

📜 License

LitGNN is released under the Apache 2.0 license. See the LICENSE file for details.