Using the Notebooks

Part 1 (2018)
1

So I understand that we are supposed to be able to recreate the notebooks from scratch. I am able to understand most of the concepts but have been struggling to recreate them from scratch. Is there anything more I could do to increase my understanding so I can recreate them from scratch?

Thanks!

2

What do you mean by “recreate”?

3

@TomLisankie As in rewrite the notebook from scratch. The fast.ai wiki said that was a good way to test my understanding of the notebooks and I was wondering what a good way to internalize the code was because I read it over, and typed it out numerous times but I still can’t recreate the notebook.

4

I think you can try to start with the simplest things then. Check the notebooks from the PyTorch tutorials. They go from the beginner to advanced level. Try to start with tensors multiplications and/or simple training loops. Also, probably you don’t need to re-implement the exactly same steps presented in the lecture. Try to setup for yourself a concrete task, like, “Create a cats/dogs classifier”, the write a small to-do list to achieve this goal, like:

  1. Download the data
  2. Prepare the data files into format expected by fastai classes
  3. Pick a pre-trained architecture
  4. Replace top layers with your own fully-connected layers
  5. Freeze the network except your top layers
  6. etc…

And then just start implementing steps required to achieve this goal using fastai library.