On the issue that modifying num_epochs in the nnUNetTrainer.py does not work

[iWangTing]

Hello, the default 1000 epochs, which is too long for my existing experimental environment to train in, in nnUNetTrainer.py Modify the number of epochs, but it doesn't work, how to deal with it, thank you！

[LambdaLi]

@iWangTing Just modifying the self.num_epochs in nnUNetTrainer.py, I succeeded by doing so, and that's not difficult. Alternatively, you can create a new Trainer that inherits the original one and then modify what you want to modify within it, all of which are based on my experience.

[ykirchhoff]

Hi @iWangTing,

changing self.num_epochs in the trainer should work. I would still recommend creating your own Trainer class, inheriting from the nnUNetTrainer as @LambdaLi suggested to avoid confusions.

Best,
Yannick

[iWangTing]

Hi， @ykirchhoff
Thank you for your reply. Although I understand your suggestion, could you please explain in detail the steps to create your own Trainer class.

Best，
Wang

[ykirchhoff]

Hi @iWangTing,

some examples of different trainer classes are given under the variants. Just follow that structure and add you own trainer somewhere down the line from variants (it might make sense to create your own folder, don't forget to include an empty __init__.py file).

Best,
Yannick