Usage

Graphtester can load, label, analyze, and test datasets. The following example shows how to load a dataset, label it, and test it with a GNN.

Loading a dataset

Graphtester package exposes a load function that can load a dataset from various formats, and convert it to a Dataset object, internal storage format of Graphtester. load function can take a dataset name, list of NetworkX or iGraph graphs, or a PyG/DGL dataset as input. The following example loads the MUTAG dataset.

from graphtester import load
dataset = load('MUTAG')

Running 1-WL or k-WL on a graph pair

Graphtester can run 1-WL or k-WL on a pair of graphs. The following example runs 1-WL on the first two graphs in the dataset.

from graphtester import (
    weisfeiler_lehman_test as wl_test,
    k_weisfeiler_lehman_test as kwl_test
)

G1, G2 = dataset.graphs[:2]
is_iso = wl_test(G1, G2)
is_iso_kwl = kwl_test(G1, G2, k=4)

Computing upper score bounds for a dataset

Graphtester can compute upper score bounds for a dataset and task. For calculating the upper score bounds associated with a specified number of layers, we utilize the congruence of Graph Neural Networks (GNNs) and graph transformers with the 1-Weisfeiler-Lehman test (1-WLE), as established in the paper.

import graphtester as gt

dataset = gt.load("ZINC")
evaluation = gt.evaluate(dataset)
print(evaluation.as_dataframe())