In trying to understand what is really going on when switching from .train() to .eval() mode, I’ve been experimenting with replacing it with the following loop within my model:
for module in self.children(): module.training = False
Is this exactly the same as calling model.eval() from outside though? A few tests with BatchNorm and Dropout seem to look good enough; but if not, what would be required to fully replace .eval?
Note: This is just for understanding, I don’t think there is any benefit in doing so