I noticed that in both the dl1 and dl2 imdb notebook there is a line like this:
learn.reg_fn = partial(seq2seq_reg, alpha=2, beta=1)
I looked back through the lessons and through all the fastai/old code, but I can’t figure out what this does. It gets passed in the kwargs to fit() in model.py, but I don’t see it getting used anywhere. I also see no reference to this property at all in fastai v1.
I have two questions:
- What is reg_fn? Is it important?
- If it is important, how do I use it in fastai v1?