Research Running Multiple Clients w/ Flower

[minaremeli]

We use Flower to simulate Federated Learning between clients. However, we don't want to start clients by hand. We will look into solutions for simulating these clients automatically. (For example, Ray.)

[VasundharaAgarwal]

Let’s use flwr.simulation (Ray) for this.

Example usage : https://github.com/adap/flower/tree/715d029422773031a14a6044e31590f6d6f75640/examples/simulation_pytorch

Instructions to run (tried and tested):

1. run git clone --depth=1 https://github.com/adap/flower.git && mv flower/examples/simulation_pytorch . && rm -rf flower && cd simulation_pytorch
2. run poetry install (Poetry is a Python package dependency manager. Can be installed with pip.)
3. run poetry shell
4. run python main.py

[minaremeli]

Yeah, works fine on my computer!

Some minor remarks:

• Did not work on first try. The problem was with the default system Python installations on Mac. Had to install Python again with pyenv. This reassured me that we need Docker so no one else suffers from random issues like this 😃
• I had to use sudo for some reason with poetry install (think this is Mac specific though, don't think it will be a problem on a Linux-based Docker image).

[minaremeli]

Also, poetry seems like a nice thing for package management, never used it before!

There might be some issues with making it work with Docker, just leaving some links here for my future self if I encounter any problems:

• https://stackoverflow.com/questions/53835198/integrating-python-poetry-with-docker
• https://github.com/python-poetry/poetry/discussions/1879
• https://medium.com/@harpalsahota/dockerizing-python-poetry-applications-1aa3acb76287