How to use the Provided Notebooks

From Deep Learning Course Wiki
Jump to: navigation, search

We have provide the jupyter notebooks and python files that Jeremy uses in the class lectures in a github repository. Note that in the videos, Jeremy may reference files being available on platform.fast.ai, but we have since switched to github and the versions on github are most up to date. If you are new to Github, here is how to get started. This is a strategy for how to get the most out of the notebooks.

Motivation

First of all, please do not just push shift-enter, shift-enter to execute through the notebooks. The test of whether you understand code is not to read it and think "yeah, that makes sense". The test of whether you understand something is whether you can build it yourself.

There are at least 3 reasons for this:

  1. Setting something aside and re-creating it forces your brain to actively recall the information, as opposed to passively reviewing. It's the difference between reading a textbook and using flash cards to quiz yourself. Active recall, unlike passive review, has been shown to be efficient in forming long-term memory.
  2. It forces you to think about what step to take next, which ensures that you are thinking carefully about the *process*.
  3. Through experimenting with inputs and outputs you build the intuitive understanding of the various components of a neural network that is critical to developing your expertise.

How to use the Provided Notebooks

  1. Read through the notebook. If everything makes sense, put it aside and create a new notebook.
  2. Now try to code the same process as we went through in class.
  3. If you get stuck at any point, you can refer to the class notebook. Find the solution to what you are stuck on. Look up the relevant documentation. Put the class notebook aside again, go back to your notebook, and try to code the solution.
  4. If you are still stuck, you can refer to the class notebook again. Do not copy and paste the needed code. Instead, type it out yourself. Check that it runs. If so, try changing the inputs, and see if that effects the outputs as you expect.
  5. Any time that you feel unsure about why a particular step is being done, or how it works, or why the outputs and inputs are what you observe (or anything else!), please ask on the forums. As I write this (week 3 of the course) there has not been a single question on the forums that has not been resolved!  :)

If the above process is easy for you, you can re-create the class notebooks with a different dataset (Look at Image Datasets for ideas).