Wasserstein Weisfeiler-Lehman Graph Kernels

This repository contains the accompanying code for the NeurIPS 2019 paper Wasserstein Weisfeiler-Lehman Graph Kernels available here. The repository contains both the package that implements the graph kernels (in src) and scripts to reproduce some of the results of the paper (in experiments).

The fork adds some converter code for novel experiments with this code.

Dependencies

WWL relies on the following dependencies:

numpy
scikit-learn
POT
cython

Converter Code relies on

networkx

Installation

To replay the experiments, clone this repo.

Usage

The WWL package contains functions to generate a n x n kernel matrix between a set of n graphs.

The API also allows the user to directly call the different steps described in the paper, namely:

generate the embeddings for the nodes of both discretely labelled and continuously attributed graphs,
compute the pairwise distance between a set of graphs

Please refer to the src README for detailed documentation.

Experiments

You can find some experiments in the experiments folder. These will allow you to reproduce results from the paper on 2 datasets.

Contributors

WWL is developed and maintained by members of the Machine Learning and Computational Biology Lab:

Matteo Togninalli (GitHub)
Elisabetta Ghisu (Github)
Bastian Rieck (GitHub)

The converter code and small changes introduced here are developed by

Pascal Welke (GitHub)

Citation

Please use the following BibTeX citation when using our method or comparing against it:

@InCollection{Togninalli19,
  author        = {Togninalli, Matteo and Ghisu, Elisabetta and Llinares-López, Felipe and Rieck, Bastian and Borgwardt, Karsten},
  title         = {Wasserstein Weisfeiler--Lehman Graph Kernels},
  booktitle     = {Advances in Neural Information Processing Systems 32},
  year          = {2019},
  pubstate      = {forthcoming},
  eprint        = {1906.01277},
  archiveprefix = {arXiv},
  author+an     = {4=highlight},
  primaryclass  = {cs.LG},
}