Hi all,
just finished lesson 1 and jumped straight to the kaggle digit recognition competition. Fastai library helped me to end up to top 7% just after one lesson and without any prior knowledge of DL. Amazing! Then I tried to tune up the learning rate little bit where I ran into problem.
In lesson one Jeremy uses lr_find() method after training the model. However, if I do so, lr_find() is completed just after 1.14% and plot is non-existent. If I look for learning before fitting, everything works fine.
I’d like to ask what causes this problem and which way is the correct one to tune up learning rate?
