Lr_find in fastai 1.0.44.dev0

Part 1 (2019)
1

In the past (about 1 month ago), the unet notebook example works just fine. i.e. this would work:

learn = unet_learner(data, models.resnet18, metrics=metrics, wd=wd)

lr_find(learn)

learn.recorder.plot()

lr=3e-4

learn.fit_one_cycle(10, slice(lr), pct_start=0.9)

But after I updated to v1.0.44.dev0, the train loss would not go down, and I have to redefine learn before running fit_one_circle. Anyone knows why?

2

We just pushed a fix for this, there was a bug in lr_find

3 Likes
3

Thank you!