Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

结果无法复现 #1

Open
vitaminzl opened this issue Aug 20, 2024 · 3 comments
Open

结果无法复现 #1

vitaminzl opened this issue Aug 20, 2024 · 3 comments

Comments

@vitaminzl
Copy link

vitaminzl commented Aug 20, 2024

你好,首先非常感谢你提出的工作。

但我在跑你的代码时,和论文的结果相差较大。

比如 cora 数据集的结果如下

 Final MICRO-F1(%): val = 80.10 ± 0.11, test = 82.53 ± 0.44                                                                                                                                                         
 Final MACRO-F1(%): val = 78.75 ± 0.11, test = 81.90 ± 0.2

而论文所给出的数据是 84.62 ± 0.37 和 82.97 ± 0.92

再比如 citeseer 数据集的结果如下

 Final MICRO-F1(%): val = 68.72 ± 0.38, test = 70.70 ± 0.70
 Final MACRO-F1(%): val = 61.38 ± 0.69, test = 64.56 ± 1.14

而论文给出的数据是 73.60 ± 0.16 和 68.11 ± 0.48,差了有 4 个点

我使用的是代码给的超参数,并未修改。

请问一下可能会是什么原因呢?

@Newiz430
Copy link
Owner

感谢关注 ❤️

能否提供一下代码的运行环境(Windows/Linux,python 包的版本,CUDA 版本等等)?我会尝试复现问题。

可以运行 run.ipynb 的第一个 block 对照一下 python 包的版本是否一致。

@vitaminzl
Copy link
Author

python 版本为 3.11.9,操作系统为 Ubuntu 20.04,CUDA 版本为 12.1
以下是相关 python 包的版本

Package numpy, version 1.26.4
Package sklearn, version 1.5.0
Package torch, version 2.3.0+cu121
Package torch_geometric, version 2.5.3
Package torch_cluster, version 1.6.3+pt23cu121
Package torch_sparse, version 0.6.18+pt23cu121
Package ogb, version 1.3.6

也许和环境、包版本相关,但不清楚具体的原因

@Newiz430
Copy link
Owner

可以尝试用CUDA 11.3运行。我们目前暂时没有CUDA 12.1的环境测试代码,后续会尝试复现并给出解决方案。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants