FLgo学习 - 技术栈

FLgo学习

Vin0sen2025-06-04 9:14

安装：pip install flgo numba

作者：https://zhuanlan.zhihu.com/p/614218973

quick-start

python 复制代码

import flgo
import flgo.benchmark.mnist_classification as mnist
import flgo.benchmark.partition as fbp
import flgo.algorithm.fedavg as fedavg

# Line 1: Create a typical federated learning task
flgo.gen_task_by_(mnist, fbp.IIDPartitioner(num_clients=5), './my_task')
# Line 2: Running FedAvg on this task
fedavg_runner = flgo.init('./my_task', fedavg, {'gpu': [0,], 'num_rounds':20, 'num_epochs': 1})
# Line 3: Start Training
fedavg_runner.run()

# Show Training Result
import flgo.experiment.analyzer as fea
# Create the analysis plan
analysis_plan = {
    'Selector':{'task': './my_task', 'header':['fedavg',], },
    'Painter':{'Curve':[{'args':{'x':'communication_round', 'y':'val_loss'}}]},
}

fea.show(analysis_plan)

数据异构

diversity 类型
Dirichlet 分布

讨论两种数据异构类型，对训练准确率的影响

不知道为什么改 num_clients 就会报错，难道只能是100？

有时间再研究下代码