Lesson2- almost flat lr curve plot when classifying with my own dataset of images

Dear all,

I am a new user of the fastai library. After watching the second week’s video I wanted to classify my own dataset of 29 classes of signal images. Yet, after following the instructions and plotting the lr curve to decide the best interval of lr values I got the following graph:

Which is not exactly what I was waiting for… there doesn’t seem any steep really…
what do you advise me to do in this case?

Thank you!

1 Like

Hi remapears hope your well!
If it were my graph I would do the following.

  1. Try values between 1e-03 and 1e-02
  2. Check my images in my dataset are in the right classes and the dataset is clean eg. no incorrect images.
  3. I would experiment by doing a few or more epochs of training with different learning rates…

Cheers mrfabulous1 :smiley::smiley:

1 Like

Thank you @mrfabulous1 for such a response! WIll do! also, specifying the interval of lr to plot gives a better view of the curve.
Hope you are having a fab day! :smiley:

1 Like