Understanding lr_find

I’m right at the beginning on Computer vision | fastai. N.B. fastai V2.

I’ve got as far as “Classifying breeds”

The output in response to:
learn.lr_find()
is “SuggestedLRs(lr_min=0.010000000149011612, lr_steep=0.0063095735386013985)” and a graph.

The text then says, "Let’s use 3e-3 " - sorry, I don’t understand how to get 3e-3 from the suggested LRs.

As I understand it 3e-3 is notation for exponential; correct me if I’m wrong but -3 means move the decimal point 3 places to the left, so 3 becomes 0.003. Neither lr_min nor lr_steep seem to be close to this number.

No one else has asked this question so I guess I’m way off base (and am asking a stupid question), but I don’t feel I can move on until I understand.

AFAIK 1e-3 is most used value for LR in many of the experimentation and its been found to be good. try both 1 e-3 and suggested LRs also.

1 Like