GraphMAE is a self-supervised learning framework for graphs, offering a generative alternative to contrastive methods. It targets researchers and practitioners in graph representation learning, aiming to achieve competitive or superior performance on node classification, graph classification, and molecular property prediction tasks.

How It Works

GraphMAE employs a masked autoencoder approach, similar to BERT for text. It masks a portion of the input graph's structural or feature information and trains an autoencoder to reconstruct the original graph. This generative pre-training allows the model to learn robust graph representations without relying on negative sampling or manual augmentation strategies common in contrastive methods.

Quick Start & Requirements

• Install: pip install torch dgl pyyaml (ensure PyTorch version >= 1.9.0, DGL >= 0.7.2)
• Prerequisites: Python >= 3.7.
• Execution: Run provided bash scripts (e.g., sh scripts/run_transductive.sh) or Python scripts (e.g., python main_transductive.py).
• Datasets: Supported datasets are automatically downloaded via DGL.
• Links: Chinese Blog, English Blog, GraphMAE2

Highlighted Details

• Achieves state-of-the-art or competitive results across multiple benchmarks.
• Offers both transductive and inductive learning settings.
• Supports various graph neural network architectures (GAT, GIN).
• Includes implementations for node classification, graph classification, and molecular property prediction.

Maintenance & Community

• The project is associated with Tsinghua University (THUDM).
• A successor, GraphMAE2, has been published with code available.
• A PyG implementation is available on a separate branch.

Licensing & Compatibility

• The repository does not explicitly state a license in the README.

Limitations & Caveats

• The README does not specify a license, which may impact commercial use or integration into closed-source projects.
• The project is from 2022, and a newer version (GraphMAE2) is available, suggesting potential for deprecation or breaking changes in this version.