18460 Project

Installation

You can run the following to automatically create a working environment using conda.

conda create -n opt python=3.11
conda activate opt
conda install jupyter notebook

pip install -r requirements.txt

Running

The easiest way to run the project is to open main.ipynb using jupyter in the conda environment.

jupyter notebook

Minimal usage

import numpy as np
from server import Server
from tasks import lrt

task = lrt.LinearRegressionTask() # will use default configuration
task.visualize()

server = Server(task)
server.connect_clients()

problem = task.get_problem()

k = 0
last_cost = float('infinity')
while k < problem.max_iter and server.delta > problem.tol:
    current_cost = problem.loss(server.consensus.flatten(), 
                                task.dataset, problem.hyper_parameters)
    server.run_iteration(k+1)
    
    if np.linalg.norm(current_cost - last_cost) < problem.ctol:
        last_cost = current_cost
        break
        
    last_cost = current_cost
    k = k + 1

if k >= problem.max_iter and server.delta > problem.tol:
    raise Exception("Did not converge")
    
solution = server.consensus.flatten()
task.visualize_solution(solution)

References

FedHybrid Paper: https://ieeexplore.ieee.org/document/10026496