Python implementation of the Weisfeiler-Lehman Graph Kernels (WLKs) method. This package is an alternative to FastWLK, GraKel, and other implementations of the WLKs method.

By design, PyWLGK is installable from PyPI and Anaconda. To install the package with pip, run the following command:

pip install pywlgk

or with conda (alternatively with mamba by replacing conda with mamba

conda install pywlgk

PyWLGK is designed to be easy to use. The following example shows how to use PyWLGK to compute the WLKs kernel between two graphs.

from pywlgk import wlk
import numpy as np

adjs = np.random.randint(0, 1, size=(2, 10, 10))
adjs = np.array(adjs + adjs.transpose(0, 2, 1), dtype=np.int32)
labels = np.ones((2, 10), dtype=np.int32)
wlk(adjs, labels, k=4)

PyWLGK takes as input a stack of adjacency matrices (adjs) and a stack of node labels (labels). The adjacency matrices must be symmetric, whereas the labels can have any type. Additionally, one can specify a k to control how many iterations of the kernel will be computed.