Read the docs | Paper | Installation | Citing Begin

Overview

BeGin is an easy and fool-proof framework for graph continual learning.

Our framework BeGin has the following advantages:

• BeGin is easy-to-use. It is easily extended since it is modularized with reusable modules for data processing, algorithm design, training, and evaluation.

• BeGin is fool-proof by completely separating the evaluation module from the learning part, where users implement their own graph CL methods, in order to eliminate potential mistakes in evaluation.

• BeGin provides 25 benchmark scenarios for graph from 15 real-world datasets, which cover 12 combinations of the incremental settings and the levels of problem. In addition, BeGin provides various basic evaluation metrics for measuring the performances and final evalution metrics designed for continual learning.

• To the best of our knowledge, we are the first to apply and evaluate parameter-isolation-based methods to graph CL.

Installation

You can install BeGin with the following command:

pip install -e .

Before running the command, we strongly recommend installing the proper version of PyTorch, DGL, and torch-scatter depending on your CUDA version.

Dependencies

• torch>=1.8.1
• dgl>=0.6.1
• torch-scatter>=2.0.6
• torch-sparse>=0.6.9
• torch-geometric>=2.0.4
• ogb>=1.3.4
• dgl-lifesci==0.3.0
• rdkit-pypi>=2022.9.1

For running some algorithms, you may need the following additional packages:

• quadprog
• cvxpy
• qpth

Package Usage

The tutorial and documents of BeGin are available at Here.

We also provide some running examples in examples directory.

Citing BeGin

If you use this framework as part of any published research, please consider acknowledging our paper.

@article{ko2022begin,
  title={BeGin: Extensive Benchmark Scenarios and An Easy-to-use Framework for Graph Continual Learning},
  author={Ko, Jihoon and Kang, Shinhwan and Shin, Kijung},
  journal={arXiv preprint arXiv:2211.14568},
  year={2022}
}