Part 1, Lesson 4 Rossmann: Learning Rate Plot

I have a strange learning rate plot that doesnt seem right, and I am not sure where to start troubleshooting.

My learning rate plot is strange, I am assuming I have some sort of error in my dataset (lr plot attached). The inflection point of the learning curve is about 9e-4, and there is a real noisy area between 10e-3 and 10e-4. Whereas Jeremy’s curve i think indicated a good LR at 1e-3 or 1e-2.

LRPLOT%20CAP

Never mind Ran a conda env update and a git pull of fast.ai and the lr cleared up, also some warnings. Moral of story: Always be up to date.

image