[Paper Review] DIG: A Turnkey Library for Diving into Graph Deep Learning Research

Meng Liu, Youzhi Luo+14 more|arXiv (Cornell University)|Mar 23, 2021

Advanced Graph Neural NetworksComputer Science56 references57 citations

TL;DR

DIG is a turnkey Python library that provides unified data interfaces, algorithms, and metrics for graph generation, self-supervised learning, explainability, and 3D graphs to facilitate graph deep learning research and benchmarking.

ABSTRACT

Although there exist several libraries for deep learning on graphs, they are aiming at implementing basic operations for graph deep learning. In the research community, implementing and benchmarking various advanced tasks are still painful and time-consuming with existing libraries. To facilitate graph deep learning research, we introduce DIG: Dive into Graphs, a turnkey library that provides a unified testbed for higher level, research-oriented graph deep learning tasks. Currently, we consider graph generation, self-supervised learning on graphs, explainability of graph neural networks, and deep learning on 3D graphs. For each direction, we provide unified implementations of data interfaces, common algorithms, and evaluation metrics. Altogether, DIG is an extensible, open-source, and turnkey library for researchers to develop new methods and effortlessly compare with common baselines using widely used datasets and evaluation metrics. Source code is available at https://github.com/divelab/DIG.

Motivation & Objective

• Address the pain of implementing and benchmarking advanced graph DL tasks beyond basic node/graph classification.
• Provide a unified, extensible testbed for graph generation, self-supervised learning, explainability, and 3D graph learning.
• Offer standardized data interfaces, algorithms, and evaluation metrics to enable easy method development and comparison.
• Encourage community contributions and easy extension with widely used datasets and metrics.

Proposed method

• Unified APIs for data interfaces, common algorithms, and evaluation metrics per direction.
• Integration with PyTorch, PyG, and RDKit to support graph and molecule operations.
• Implementation of four research directions: graph generation, self-supervised learning on graphs, explainability, and 3D graph learning.
• Support for widely used datasets (e.g., QM9, ZINC250k, MOSES, TUDataset) and standard evaluation metrics.
• Extensible and modular design to easily add new datasets, algorithms, and metrics.
• Quality assurance through continuous integration and documentation.
• Open-source under the GNU GPLv3 license.

Experimental results

Research questions

• RQ1 How can a unified, extensible testbed accelerate development and benchmarking of graph DL methods across multiple research directions?
• RQ2 Can researchers easily adopt DIG to implement, compare, and reproduce experiments on graph generation, self-supervised learning, explainability, and 3D graphs?
• RQ3 What design principles enable seamless extension with new datasets, algorithms, and metrics in graph deep learning?
• RQ4 How well do unified interfaces and benchmarks align with widely used datasets and evaluation practices in the graph DL community?

Key findings

• DIG provides unified and extensible implementations of data interfaces, common algorithms, and evaluation metrics for graph generation, self-supervised learning, explainability, and 3D graphs.
• The library includes 18 algorithms across four directions and leverages Python, PyTorch, PyG, and RDKit for graph and molecule operations.
• DIG emphasizes reproducibility and quality with continuous integration, online documentation, and a GPLv3 license to foster community contributions.
• Benchmarks and benchmark examples are provided to reproduce reported results within reasonable differences.
• DIG serves as a turnkey platform for implementing new methods and conducting empirical comparisons with baselines on widely used datasets and metrics.