Pets Breed data - High error rate on the Lesson notebook L05

Hi everyone,

I tried to run the same notebook provided for the Lesson 05 with Pet Breeds. Compared to the results, when I run it on Colab notebook here: https://colab.research.google.com/github/fastai/fastbook/blob/master/05_pet_breeds.ipynb

I am getting very high error rate, while running learn.fine_tune(2) on the cnn_learner - something like starting with 30%, and settling for 25%.

Can someone please confirm the same? Any suggestions why this is the case? Is it due to a different version of fastai now?

Thanks in advance.

Sam

Hi,

Try training for some more epocs
learn.fine_tune(10)

Also, please validate the input data
dls.show_batch(figsize=(9, 9))

Hi Sam,

It has to do with a recent update around 2.2.7, check out the conversation here: Performance degradation between fastai 2.2.5 and 2.2.7

Warm regards,

Ari

Thanks Edward and Ari for your response. I can now follow what went wrong and am tracking the progress on the thread shared by @jankelowitz.

Sam

1 Like