关于noniid-#label2数据划分每次运行程序划分的结果都不一样的问题 about niid-bench HOT 2 CLOSED

Thanks for your question.

Yes, we also find this is a problem. After examining the code, we find that it is because we use random.randint(). This seems not following the np.random.seed(). To solve it, we need to set seed for all these functions.

np.random.seed()
torch.manual_seed()
random.seed()

We have added them in experiments.py. After that, the partition of noniid-#label2 is fixed. You can experiment on exactly the same partition then.

Thanks again for pointing it out and help improving our project. Please let us know if there are further questions.

from niid-bench.

感谢回复，问题已经得到解决

from niid-bench.