Hi there!
Thanks for an awesome course! I’ve tried to run it for my own dataset I’ve created (dragonflies vs butterflies) and got incredible 0.9869 accuracies from the first try:
https://github.com/vborees/fastai/blob/master/lesson1/model_dragonfly_butterfly/lesson1-dragonfly-butterfly.ipynb
As you can see, I got two error messages because of so incredible results.
- Cell 26, when tried to show Incorrectly classified, a ValueError occurred because there are less than 4 incorrectly classified images and np.random.choice could not get 4 out of 2.
- Cell 35 gives ‘<Figure size 1152x576 with 0 Axes>’, because there all butterflies have been classified correctly. Fix: if there are no images, do not plot, but just print a message.
There are minor fixes for that in the pull request I created: https://github.com/fastai/fastai/pull/744