search

TsingZ0 / PFLlib

37 traditional FL (tFL) or personalized FL (pFL) algorithms, 3 scenarios, and 20 datasets.
GNU General Public License v2.0
1.35k stars 283 forks source link

Hyperparameter settings for some models #183

Closed YouthBL closed 4 months ago

YouthBL commented 4 months ago

Hi, Jianqing, thank you very much for your open-source project, it's great work and helped me a lot in my studies. Recently, I ran the FedRep and FedALA models on the Cifar10 dataset under the Pathological heterogeneous setting， the hyperparameters I used were ''python main.py -lr 0.005 -lbs 10 -nc 20 -jr 1 -nb 10 -data Cifar10 -m cnn -algo FedXXX -gr 2000 -did 0''. The results are 91.14 for FedRep and 91.75 for FedALA. There is a gap of about 0.7% compared to what is reported in Table 2 of the FedALA paper. So I want to ask for help on how to tune my hyperparameters. Thanks again for the great work on the project.

TsingZ0 commented 4 months ago

Thank you for your attention! As mentioned in the README.md, this project updates gradually, and some basic settings may also change, so the gap of 0.7% is negligible in this case. I typically use grid search to tune hyperparameters. #136 also provides some choices.

YouthBL commented 4 months ago

I got it. Thanks for your reply.