loss is nan, accuracy is quite low, and bug

[Geeks-Z]

Thanks for your work, I have some problems with running the default params:

[Taoxiaohei]

hello, i meet a same problem like you, do you deal with it?

[Threepointso]

hello i met the same problem do you deal with it? Can we communicate about this?

[CCDxibo]

hello i met the same problem do you deal with it? Can we communicate about this?

[Taoxiaohei]

@CCDxibo let me guess, do you use 5090 to make it?

[CCDxibo]

@Taoxiaohei I did not use 5090, and I have now preliminarily resolved the above issue. It seems that the error was caused by version issues with libraries such as numpy, scipy, pandas, etc. After resetting the conda environment according to the requirements file configuration, the data seems to be normal now
You can also check the closed issues, which provides some solutions to version issues

[Threepointso]

是这样的 最后显示是conda环境.部分有问题

…

---Original--- From: ***@***.***> Date: Fri, Aug 15, 2025 14:46 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [KU-VGI/PriViLege] loss is nan, accuracy is quite low, and bug(Issue #13) Taoxiaohei left a comment (KU-VGI/PriViLege#13) @CCDxibo let me guess, do you use 5090 to make it? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[CCDxibo]

是这样的 最后显示是conda环境.部分有问题
…

嗯嗯是的，感谢！

[g8364]

@Taoxiaohei I did not use 5090, and I have now preliminarily resolved the above issue. It seems that the error was caused by version issues with libraries such as numpy, scipy, pandas, etc. After resetting the conda environment according to the requirements file configuration, the data seems to be normal now You can also check the closed issues, which provides some solutions to version issues
Hello, i met the same problem do you deal with it? Have you successfully run it? Is the result reasonable?

[CCDxibo]

@Taoxiaohei I did not use 5090, and I have now preliminarily resolved the above issue. It seems that the error was caused by version issues with libraries such as numpy, scipy, pandas, etc. After resetting the conda environment according to the requirements file configuration, the data seems to be normal now You can also check the closed issues, which provides some solutions to version issues
Hello, i met the same problem do you deal with it? Have you successfully run it? Is the result reasonable?

Yes, I have resolved the above issue and successfully run the code. The problem lies in the version of the runtime environment provided by the author, and you may need to manually install some libraries yourself. The final accuracy of the operation is about 1% lower than the results given in the paper, but still within an acceptable range. I think this may be due to differences in seed settings, parameter configurations, hardware, and other aspects

[g8364]

Thank you for your reply. You've done an excellent job! I'm new to this field and still have a lot to learn. I really appreciate your reply.

…

---Original--- From: ***@***.***> Date: Fri, Oct 17, 2025 15:30 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [KU-VGI/PriViLege] loss is nan, accuracy is quite low, and bug(Issue #13) CCDxibo left a comment (KU-VGI/PriViLege#13) @Taoxiaohei I did not use 5090, and I have now preliminarily resolved the above issue. It seems that the error was caused by version issues with libraries such as numpy, scipy, pandas, etc. After resetting the conda environment according to the requirements file configuration, the data seems to be normal now You can also check the closed issues, which provides some solutions to version issues Hello, i met the same problem do you deal with it? Have you successfully run it? Is the result reasonable? Yes, I have resolved the above issue and successfully run the code. The problem lies in the version of the runtime environment provided by the author, and you may need to manually install some libraries yourself. The final accuracy of the operation is about 1% lower than the results given in the paper, but still within an acceptable range. I think this may be due to differences in seed settings, parameter configurations, hardware, and other aspects — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>