Awesome

KAN4Graph

Implementation of Kolmogorov-Arnold Network (KAN) for Graphs. Any communications, collaborations, issues, PRs are welcomed. The contributors will be listed here. Please contact yueliu19990731@163.com. If you find this repository useful to your research or work, it is really appreciate to star this repository. :heart:

<details> <summary>Table of Contents</summary> <ol> <li><a href="#Usage">Usage</a></li> <li><a href="#acknowledgement">Acknowledgement</a></li> </ol> </details>

Usage

Requirements

KAN4Graph is implemented with Python3.8.16 and 1 NVIDIA Tesla V100 SXM2 16 GB

Python package information is summarized in requirements.txt:

torch==1.7.1
tqdm==4.59.0
numpy==1.19.2
munkres==1.1.4
scikit_learn==1.2.0

Datasets

Dataset	Type	# Nodes	# Feature Dimensions	# Edges	# Classes
BAT	Attribute Graph	131	81	1038	4
UAT	Attribute Graph	1,190	239	13,599	4
EAT	Attribute Graph	399	203	5,994	4

still updating...

Quick Start

clone this repository and change directory to KAN4Graph

git clone https://github.com/yueliu1999/KAN4Graph.git
cd ./KAN4Graph

run codes

python train.py

Results

Dataset	Metric	Score
BAT	ACC	77.86
	NMI	54.48
	ARI	52.33
	F1	77.34
UAT	ACC	57.05
	NMI	25.49
	ARI	24.97
	F1	55.80
EAT	ACC	57.87
	NMI	34.16
	ARI	27.52
	F1	58.09

still updating...

Acknowledgements

Our code are partly based on the following GitHub repository. Thanks for their awesome works.

pykan: the official implement of KAN.
fast-kan: the implement of KAN (fast version).
Awesome Deep Graph Clustering: a collection of deep graph clustering (papers, codes, and datasets).
SCGC: the official implement of Simple Contrastive Graph Clustering (SCGC) model.

Contributors