Low-dimensional statistical manifold embedding of directed graphs (ICLR 2020)

by Thorben Funke, Tian Guo, Alen Lancic, and Nino Antulov-Fantulin

Link to publication: ICLR 2020 Poster, arXiv 1905.10227, OpenReview

Install requirements

If you want to execute our code, you should have the following packages installed:

• tensorflow >=2.0
• networkx >=2.3
• numpy >=1.16
• scikit-learn >= 0.21.3

For Monte-Carlo integration:

• tensorflow_probability >= 0.7

Executing

1. As first step, the folder and the configuration needs to be created with the following command, which creates a folder for this configuration and stores the settings in an .ini-file.

   python create_configuration.py
   

2. Execute the created configuration with

   python execution.py
   

Depending on the availability of a GPU, the code will use the specified GPU (default 1) or run on the CPU. You may need to set the parameter -d or --dataset, if the memory usage of the GPU is too high.

During execution a logfile will be created within the created folder. The results will be saved as .npz file (numpy.load to access the data) to the subfolder results.

Executing the above steps without modifications, will generate the embedding of our toy example of 25 nodes. To execute different networks, supply the path to a edge list of the same format or adjust the function read_graph_from_file of the generate_data.py file accordingly. (You can use any graph format, which NetworkX supports.)

The network files for the experiments, you can get from the KONECT - The Koblenz Network Collection project.

Reference

Funke, T., T. Guo, A. Lancic, and N. Antulov-Fantulin (2020). “Low-dimensional Statistical Manifold Embedding of Directed Graphs.” In: Proceedings of the 8th International Conference on Learning Representations. ICLR’20.

@inproceedings{funke2019statistical,
title={Low-dimensional Statistical Manifold Embedding of Directed Graphs},
author={Funke, Thorben and Guo, Tian and Lancic, Alen and Antulov-Fantulin, Nino},
booktitle={Proceedings of the 8th International Conference on Learning Representations},
series = {ICLR'20},
year={2020}
}