EarlyStopping every two epochs

[quentinblampey]

Hello,

My model is training on two different dataloaders: loader1 is used on even epoch number, and loader2 for odd numbers.

I'm using EarlyStopping, but I want it to focus only on the even number of epochs, i.e. when we are usingloader1 (thus, I don't want to consider metrics logged from loader2).
Do you have any idea how to do this?

I tried to log the loss only for loader1, but it says:

You called `self.log(loss, ...)` twice in `training_step` with different arguments. This is not allowed

I could also multiply the patience by 2, but the loss is usually lower on the loader2, so the current lower loss value is updated everything when training on the second loader.