wenzhu23333 / Federated-Learning

An open source FL implement with dataset(Femnist, Shakespeare, MNIST, Cifar-10 and Fashion-Mnist) using pytorch
GNU General Public License v3.0
115 stars 27 forks source link

nonIID下的MNIST-CNN #2

Closed b856741 closed 2 years ago

b856741 commented 2 years ago

你好 我執行python main.py --dataset mnist --num_channels 1 --model cnn --epochs 50 --gpu 0 發現testing accuracy 都非常低,維持在固定數字9.80的也有 哪裡有出錯嗎?(mlp沒發現這現象)

wenzhu23333 commented 2 years ago

调整一下learning rate,加一个参数 --lr 0.01,learning rate现在默认为1,对CNN来说太大了。

wenzhu23333 commented 2 years ago

已修正命令及默认值。