WebMILE: Democratizing Network Representation Learning at Scale

WebMILE is an easy-to-use embedding framework based on our previous works MILE [1] and DistMILE [2].

Requirements

• Python: Our web interface and MILE system are Python-based. At your convenience, please use Conda to create a Python environment with requirements.yml. Our dependency mainly includes:

  • Django
  • NumPy
  • Scikit-learn
  • SciPy
  • TensorFlow

• Docker: Required for execution of base embedding script in different languages. Please follow this link to install docker on your device.

How to start

To launch the service, use this command in your terminal:

python manage.py runserver

How to use

• First, please upload your graph data. WebMILE supports two formats: edgelist and metis [3].
• Then specify the coarsen level and embedding dimension.
• Choose your programming languages. Currently we support Python, Java, and R.
• Upload your embedding script. Both a single script and a zip file. The main script should be named embed.{py/jar/R}.
• [Required for Python and R users] Upload a dependency list. Java users should include all needed libraries in their JAR file.
• [Optional] Add arguments in the command box.
• Start running MILE. When finished, a link will be provided for downloading the embeddings.

Examples

A small real-world graph dataset and two embedding scripts in Python (DeepWalk [4]) and Java (node2vec [5]) are included in /examples/. When you write your own script, please make sure that four arguments are required:

• --input: Path of the input coarsened graph data (in the format of edgelist)
• --output: Path of the output embedding file.
• --embed-dim: Dimensonality of embeddings.
• --workers: For parallel computing.

Reference

[1] Liang, Jiongqian, Saket Gurukar, and Srinivasan Parthasarathy. "MILE: A Multi-Level Framework for Scalable Graph Embedding." Proceedings of the International AAAI Conference on Web and Social Media. Vol. 15. 2021.

[2] He, Yuntian, Saket Gurukar, Pouya Kousha, Hari Subramoni, Dhabaleswar K. Panda, and Srinivasan Parthasarathy. "DistMILE: A Distributed Multi-Level Framework for Scalable Graph Embedding." 2021 IEEE 28th International Conference on High Performance Computing, Data, and Analytics (HiPC). IEEE, 2021.

[3] Karypis, George, and Vipin Kumar. "METIS: A software package for partitioning unstructured graphs, partitioning meshes, and computing fill-reducing orderings of sparse matrices." (1997).

[4] Perozzi, Bryan, Rami Al-Rfou, and Steven Skiena. "Deepwalk: Online learning of social representations." Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining. 2014.

[5] Grover, Aditya, and Jure Leskovec. "node2vec: Scalable feature learning for networks." Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining. 2016.