UnboundLocalError: local variable 'shape_estimator' referenced before assignment about fedma HOT 2 OPEN

Hi @zarekxu, thanks a lot for your interest in our work! The shape_estimator is used to calculate the input dimension of the very first FC layer in a CNN model architecture e.g. LeNet or VGG. Based on your description, it seems there are some issues with model weight reloading in the MNIST+LeNet experiment.

Can you please provide more details on how to reproduce the issue e.g. what was launch script you used and which exact line reports that error?

Thanks again and I'm happy to provide more help on this issue.

from fedma.

Hello Authors,
I ran the vgg on Mnist and got the "shape_estimator error. I didn't change anything in the code, just wanted to recreate the results of the paper. Here is the content of the run.sh file:

python main.py --model=vgg
--dataset=cifar10
--lr=0.01
--retrain_lr=0.01
--batch-size=64
--epochs=20
--retrain_epochs=20
--n_nets=16
--partition=hetero-dir
--comm_type=fedma
--comm_round=10
--oneshot_matching=
--retrain=True

So, can you tell me what is the issue?

from fedma.