zhiqiangzhongddu / selene Goto Github PK

Thank you for your innovative work in graph clustering on heterophily graphs.
I found your work highly intriguing and insightful.
As such, I tried to reproduce your work (Table 1).
In the meanwhile, I encountered several issues and hope to raise some questions regarding your codes.

I executed your source code with the command "python main.py --data_name Texas --p_x 0.1 --p_e 0.0 --seed 0 --feat_method sp --parallel True". (hyperparameters based on your Hyperparameter Record.xlsx file)

The results are as follows:

is evaluating model...
K-means test
ACC: 49.62-5.48
NMI: 7.88-2.62
ARI: 6.37-5.21
F1: 23.16-3.63
model evaluation is done.
Evaluating X embedding...
K-means test
ACC: 47.32-5.74
NMI: 7.39-1.62
ARI: 6.29-4.44
F1: 23.45-1.53
Done
Evaluating G embedding...
K-means test
ACC: 46.56-1.62
NMI: 5.62-0.42
ARI: 4.07-1.49
F1: 24.15-0.60
Done

How to obtain the performance reported in Table 1. (Texas: ACC 65.23 / NMI 23.40 / ARI 34.21) ?

Great congratulations to your TMLR paper. Could you provide the detailed hyper-parameter settings of the proposed method or the detailed codes to reproduce the results in the original paper.