clustering reproduce question

[sakura8999]

Hello, I conducted the clustering experiments on the DBLP dataset , but the scores are lower than the results reported in your paper.
The best epoch is: 293[clustering] nmi: [.7192, .92] ari: [.768, .092]
Are there any other factors I should pay attention to besides the parameters you recommended? thank you.

[jhqi]

Hello, I conducted the clustering experiments on the DBLP dataset , but the scores are lower than the results reported in your paper. The best epoch is: 293[clustering] nmi: [.7192, .92] ari: [.768, .092] Are there any other factors I should pay attention to besides the parameters you recommended? thank you.

hi, I've also reproduced the clustering task on DBLP. I directly used the code given by the author below:
python main.py --dataset dblp --task clustering --use_cfg

and I got a quite poor result, even much lower than yours. Is there anything I need to modify on the author's code? plz

I didn't do anything to the source code. And it seems that the author has fixed the random seed in the code. But my training state is different from yours:

[sakura8999]

I Sorry, I was a little late to see the email. I didn't change the author's code when I reproduced, and I did it on the RTX4090. Maybe you might want to consider changing graphics cards. 2023-12-05 19:49:19 "qqqck18" ***@***.***> 写道： Hello, I conducted the clustering experiments on the DBLP dataset , but the scores are lower than the results reported in your paper. The best epoch is: 293[clustering] nmi: [.7192, .92] ari: [.768, .092] Are there any other factors I should pay attention to besides the parameters you recommended? thank you. hi, I've also reproduced the clustering task on DBLP. I directly used the code given by the author below: python main.py --dataset dblp --task classification --use_cfg and I got a quite poor result, even much lower than yours. Is there anything I need to modify on the author's code? plz — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you modified the open/close state.Message ID: ***@***.***>

[zyq-bupt]

I get this. It's higher than yours but still lower than that reported in the paper. I use the default parameters. Here is my device configuration, maybe it will be useful to you：4090*1, pytorch==2.1.0， cuda==12.1