FedKT
FedKT copied to clipboard
Accuracy issues on implementing
- The Accuracy constantly is 0.0 regardless of the changing hyperparameters
- When observing the vote counts, the total number of votes is always 10.
- What are the minimum and maximum limit for teacher models per party
- Is it necessary that the public dataset be the same as the private dataset?
- In case of lack of public dataset, how will model wor output.txt k?
Hi ,
- In your script, the parameters
epoch
,stu_epochs
,n_teacher_each_partition
,n_partition
are too low. Can you try the provided script to see whether the accuracy is 0.0? - The total number of votes is equal to the number of models.
- THere is no limit for the number of teacher models per party. However, the performance may vary a lot with a different number of models.
- The public dataset is from the same source as the private dataset (e.g., all are from MNIST) in our experiments. The performance may drop a lot if using a different public dataset.
- FedKT does not work if there is no public dataset currently.
I changed them to the default values. But still no difference. Also, the experiment generates log files, but there is nothing in them. All I get are 0KB files with date and time of the experiment
I changed them to the default values. But still no difference. Also, the experiment generates log files, but there is nothing in them. All I get are 0KB files with date and time of the experiment
Hello, I also have this problem now, I would like to know now this problem you have solved?
Hi @Cmnotjx and @ashwiniyer176 ,
What is the operating system you used? I can successfully run the scripts and generate non-empty logs in a Linux system. There may be compatibility issues on Windows.
Hi @Cmnotjx and @ashwiniyer176 ,
What is the operating system you used? I can successfully run the scripts and generate non-empty logs in a Linux system. There may be compatibility issues on Windows.
Thank you very much for your reply, I am using a windows system, I will use a Linux system to try