Description
This GitHub repository, chennnM/GCNII, offers a PyTorch implementation of the "Simple and Deep Graph Convolutional Networks" (GCNII) research paper. The project aims to provide researchers and developers with a readily accessible and functional codebase to experiment with and build upon GCNII.
Read More
The implementation supports various graph datasets, including benchmark sets like Cora, Citeseer, and Pubmed, as well as additional datasets such as Chameleon, Cornell, Texas, and Wisconsin. It allows for the replication of both semi-supervised and full-supervised learning settings, mirroring the experimental setups described in related research like GCN and Geom-GCN. The repository also includes scripts for running these experiments, making it straightforward to reproduce reported results.
Key components of the implementation include the core model architecture, data loading and preprocessing utilities, and training scripts. The project details its dependencies, such as CUDA, Python, PyTorch, NetworkX, and scikit-learn, ensuring users can set up their environment correctly. Furthermore, a reference implementation within the PyG (PyTorch Geometric) folder is provided for those who prefer or require integration with the PyTorch Geometric library, with specific version requirements for torch-geometric and ogb.
The repository also presents a summary of testing accuracy results across different datasets and settings, offering a benchmark for users to compare their own implementations or findings. The usage instructions are clear, guiding users on how to execute the provided shell scripts for semi-supervised, full-supervised, and inductive PPI experiments. This makes the GCNII model accessible for practical application in graph-based machine learning tasks.
The project is actively maintained, with recent updates to the README file. It highlights the collaborative nature of open-source development, with contributions and forks indicating community engagement. The primary language used is Python, with a small percentage of Shell scripting for utility purposes. This resource is valuable for anyone interested in advanced graph neural networks and their applications.
- CategoryMachine Learning Platforms
- Pricing ModelOpen source
- DeploymentCloud, Self-hosted
- PlatformsLinux, Windows
GCNII's Core Features
PyTorch implementation of GCNII model
Supports semi-supervised learning
Supports full-supervised learning
Includes benchmark datasets (Cora, Citeseer, Pubmed)
Includes additional datasets (Chameleon, Cornell, Texas, Wisconsin)
Provides scripts for replicating experimental results
Offers a PyTorch Geometric reference implementation
Details dependencies for environment setup
Summarizes testing accuracy results
Includes usage instructions for experiments
Codebase written primarily in Python
Getting Started with GCNII
Set up environment: Install CUDA, Python 3.6.9, PyTorch 1.3.1, NetworkX 2.1, scikit-learn.
Download datasets: Place data in the 'data' and 'newdata' folders.
Replicate semi-supervised results: Run `sh semi.sh`.
Replicate full-supervised results: Run `sh full.sh`.
Replicate inductive PPI results: Run `sh ppi.sh`.
Integrate PyG implementation: Ensure torch-geometric >= 1.5.0 and ogb >= 1.2.0 are installed.
Run PyG examples: Execute `python cora.py` or `python arxiv.py`.
GCNII's Use Cases
- Graph Neural Network Research
- Reproducing Research Results
- Graph-based Machine Learning
- Model Benchmarking
- Educational Tool
- Inductive Learning on Graphs
