How to get consistent order of spawned clients between multiple runs with Flower's simulation?

[philip-s15]

What is your question?

Hi there!

I am using Flower for my machine learning research project, so reproducibility is quite important to me. I have noticed that when comparing multiple Flower simulation runs, the order in which the virtual clients are spawned is not consistent between the runs, even though I manually set all the libraries' seeds to a consistent value.

My question is whether it is possible to ensure that the order in which clients are spawned becomes consistent between multiple runs with the same code. I have tried looking at the Ray documentation, but I could not find anything related to manually setting a seed.

Thanks in advance!

[edogab33]

same problem here

[artsasse]

struggling with that problem too

[WilliamLindskog]

Hi @philip-s15, thanks for raising this. Is it the order that they are spawned or is it more what dataset is allocated to a specific client e.g. part_a --> client_1, part_b --> client_2 and so forth?