AttributeError: 'FedavgServer' object has no attribute 'lr_scheduler'

[vaseline555]

AttributeError: 'FedavgServer' object has no attribute 'lr_scheduler'

[vaseline555]

Dear @Jinfeng-Xu ,
Thank you for your kind words!

Oops, I forgot to change that part when deubgging.
It is fixed in this commit.
Please pull the latest version of this repo and try again.
And please let me know if there still exists a problem!

Thank you.
고마워요~

Best,
Adam

[Jinfeng-Xu]

Many Thanks!
But still, some bugs need to be fixed.

server.update() may raise an error for unexpected type

[vaseline555]

Thank you for your report, @Jinfeng-Xu !
Could you please provide me corresponding command that causing such an error?

I presume that the error caused when you used BCELoss.
If in that case (plus if the dataset you want to use has at most 2 classes), however, you should change it into BCEWithLogitsLoss since any models provided in this repo do not have torch.nn.Sigmoid() activation in the final layer of the model (i.e., last layer of the model's classifier submodule).

FYI, it is an intended behavior even if users entered metrics requiring probability ouputs (e.g., AUROC), because it is implicitly transformed in MetricManager instance.

Hope it helps.
Thank you!

Best,
Adam

[Jinfeng-Xu]

Thank you for your guidance, @vaseline555 !

Your guess is right, it works with BCEWithLogitsLoss.
Thanks for your help!

Sincerely,
Jinfeng

[vaseline555]

Sounds great!
Hope you have a nice Friday. 😄
(불금 in Korean)

Best,
Adam