Skip to content

01 DGL

Handling nodes without natural features - Learnable node features: for

Deep Graph Library (DGL)⚓︎

  • Started as research project out of NYU Shanghai
  • Joined forces with AWS and is now developed/maintained there in collaboration with NVIDIA and the open source community
  • Differentiated by speed and scalability
    • Supports multi-GPU and multi-machine with multi-GPU
  • Documentation has nice section of tutorials for learning and there are many examples in github
  • Backend is framework agnostic

DGL 1.0 release⚓︎

dgl blog

Useful APIs and data structures⚓︎

  • Named node and edge features
    • g.ndata['x']=X
    • g.edata['a']=E
  • Graph processing
    • dgl.add_reverse_edges(g)
    • dgl.add_self_loop(g)
  • Graph querying
    • g.num_nodes()
    • g.num_edges()
    • g.has_edges_between(u,v)
    • g.in_degrees()
  • Message passing APIs
    • \(h_u=\frac{1}{N_u}\sum\limits_{V\in N_u}{x_v}\) = g.update_all(fn.copy_u('x', 'm'), fn.mean('m','h'))
  • Increasingly supporting heterogeneous graphs

PyTorch Geometric (PyG)⚓︎

  • Built on top of PyTorch
  • Developed at TU Dortmund and Stanford Universities
  • Overlap with group that runs the Open Graph Benchmark
  • Standardizes the API around defining “message” and “update” functions, along with specifying an aggregator
  • Has a large set of built-in datasets and implementations

New(er) comers⚓︎

  • Jraph
    • Comes from DeepMind
    • Built on top of JAX
  • TensorflowGNNs
    • Alpha release announced Nov 2021
    • Keras-style API
    • A stated emphasis on heterogeneous graphs

Open Graph Benchmark (ogb.stanford.edu)⚓︎

  • Larger and more realistic benchmark graph datasets
    • Node, Link and Graph property prediction tasks
  • Wrapped in a Python package for easy loading into PyGand DGL
    • Comes with a pre-defined train/val/test split
  • Also comes with built in “Evaluators” for each dataset to ensure performance is consistently measured
  • Each dataset has a leaderboard, each row is a method’s performance with paper + code
  • There’s also a “Large-Scale” version with much larger graphs

References⚓︎

  • Wang, M., Zheng, D., Ye, Z., Gan, Q., Li, M., Song, X., ... & Zhang, Z. (2019). Deep graph library: A graph-centric, highly-performant package for graph neural networks. arXivpreprint arXiv:1909.01315.
  • Fey, M., & Lenssen, J. E. (2019). Fast Graph Representation Learning with PyTorch Geometric [Computer software]. https://github.com/pyg-team/pytorch_geometric
  • Godwin, J., Keck, T., Battaglia, P., Bapst, V., Kipf, T., Li, Y., ... Sanchez-Gonzalez, A. (2020). Jraph: A library for graph neural networks in jax(Version 0.0.1.dev). Opgehaalvan http://github.com/deepmind/jraph
  • Hu, W., Fey, M., Zitnik, M., Dong, Y., Ren, H., Liu, B., ... & Leskovec, J. (2020). Open graph benchmark: Datasets for machine learning on graphs. arXivpreprint arXiv:2005.00687.