Graph Convolutional Networks (GCN) with PyTorch

This repository is for a simple implementation of Graph Convolutional Networks. The original paper of GCN is here.

Datasets

In this repository, there are 2 datasets provided to train the GCN model:

- Citeseer
- Cora

The two datasets can be downloaded in here. The downloaded datasets should be unziped and restored into a directory named datasets. The following tree shows the path to each files in the datasets:

datasets
├── citeseer
    ├── citeseer.cites
    ├── citeseer.content
    └── README
└── cora
    ├── cora.cites
    ├── cora.content
    └── README

Install Dependencies

1. Install Python 3.

2. Install the Python packages in requirements.txt. If you are using a virtual environment for Python package management, you can install all python packages needed by using the following bash command:

   $ pip install -r requirements.txt

3. Install PyTorch. The version of PyTorch should be greater or equal than 1.7.0. This repository provides the CUDA usage.

Training

1. Modify config.json as your machine setting. The following explanations are for understanding train_config of config.json:

   • num_epochs: The number of epochs of the training process. Default: 200
   • learning_rate: The learning of the optimizer for the training process. Default: 0.01

2. Execute training process by train.py. An example of usage for train.py are following:

   $ python train.py --dataset_name=cora

   The following bash command will help you:

   $ python train.py -h

Training Results

Training Configuration

Dataset	Configurations
Citeseer	num_epochs: 200, learning_rate: 0.01
Cora	num_epochs: 200, learning_rate: 0.01

Performances

Dataset	Maximum Test ACC (%)	Hyperparameters
Citeseer	65.11 ± 0.77	H: 16, num_layers: 2, dropout: 0.5, regularization: 0.0005
Cora	81.75 ± 0.36	H: 16, num_layers: 2, dropout: 0.5, regularization: 0.0005

Recent Works

• Initially complemented the implementations.

Future Works

• Refactoring.
• Write docstrings.

References

• Semi-supervised Classification with Graph Convolutional Networks
• Revisiting Semi-Supervised Learning with Graph Embeddings