Fastai v2 equivalent for this v1 code

fastai users
#1

If in the v1 I had some code like below
lr = 1.2e-3
learn.fit(lr,15)
lr_lst=np.array([lr/100,lr/10,lr])
# differential
learn.fit(lr_lst/4, 3, cycle_len=1, cycle_mult=2)
# fine tune
learn.fit(lr_lst/10,1,cycle_len=5,use_clr=(10,32))
what would be the v2 equivalent of the above ?

PS: Sorry about the bad formatting.

#2

Current version on github is ‘1.0.41.dev0’

What do you mean by “V2”?

#3

The old fastai codebase does not contain many of the above options and looks like it has been superseeded by calls such as fit_one_cycle. The fit_one_cycle() call also does not contain some of the above options such as cycle_mult and use_clr.

I was wondering how I would reproduce the above code using the new fit_one_cycle and fit calls.

Thanks.

#4

I think what you mean is “Fastai v1 equivalent of this v0.7 code”

#5

Right, thanks for the correction. Is there any equivalent notebooks out there?

#6

There are 2 places you can look at the moment:

  1. official fast.ai v1 documentation here and
  2. notebooks for Deep Learning for Coders v3 here

It may not be 100% equivalent to what you’re trying to do, but will get the job done.