mpiexec -n 1 python server.py : -n 10 python client.py

--data_pattern (int)

options:

• 0: IID
• 1-4: non-IID
• 5~7: Dirichlet($\alpha$, $n$). $\alpha \in (0, +\infty) $ is the distribution parameter, n is the number of virtual workers. The lower the value of $\alpha$, the greater the degree of skewedness. The file containing data partitions (.npy) is saved in ./data_partition/
  • 5: Dir(1.0, n)
  • 6: Dir(0.5, n)
  • 7: Dir(0.1, n)

--dataset_type (str)

options: image100, CIFAR10, UCIHAR, SPEECH

--epoch (int)

80 for UCIHAR, 500 for IMAGE100, 250 for CIFAR10 and SPEECH

--worker_num (int)

the number of virtual workers, default 10

Run on UCIHAR with Dirichlet distribution parameterized by 0.1, across 10 clients for 80 epoch, with 10 clients being active each epoch

mpiexec -n 1 python server.py --data_pattern 7 --dataset_type UCIHAR --epoch 80 --worker_num 10 : -n 10 python client.py