Code missing

[MukundPatel13]

Code is missing for generating the different graph show in the published arcticle "Communication-efficient federated continual learning for distributed learning system with Non-IID data".
I would highly appreciate if this can be added to the library. Also please add instructions for how to run the code and in which order

[zhangzhao156]

Most of the graphs in the paper are ploted by Origin (a softsware) not matplotlib. You can save the runing results of the python files and then plot them in Origin.
You can just run main_minist_*.py, fedsi and cfedsi are the proposed methods in the paper, fedprox, fedewc and fedavg are the compared methods.

[MukundPatel13]

Thanks for coming back. Please advise how can I do it using matplotlib only. Also I have other question I ran cfedsi with IID data and changing no other parameter in the program the final result after 50 iterations I got :
Test set: Average loss: 0.0457
Accuracy: 7579/10000 75.79
Testing accuracy: 75.79
While in the paper you mentioned that CfedSI has accuracy of 94.91 for MNIST dataset. PLease advise the required parameter change for each run .

[zhangzhao156]

You can save the experimental results as the logging files, and then use matplotlib to write the graphs based on the results in the logging file, e.g., nohup python main_mnist_cfedsi.py > ./iid_mnist_cfedsi.log 2>&1 &
def read_log(filename):
test_loss = []
test_acc = []
with open(filename,'r') as file_to_read:
while True:
lines = file_to_read.readline()
if not lines:
break
item = [i for i in lines.split()]
if (len(item) == 5) and (item[0]=='Test'):
test_loss.append(float(item[4]))
return test_loss

You can tune the parameters of training setttings to imporve the performances, e.g., lamda, rho, s, learning rate.

[MukundPatel13]

could you please add the fashion Mnist program for FedAvg , Cfedsi. I tried changing the code but its not working for me