Refactored Fast.ai Pt2 2019 with exported .py files all in a centralised location, exported all notebooks as scripts so that source code goto definition can be used in an IDE.
(I found it super helpful to step through the code in a debugger).
At lesson 12 stage I found it too hard to keep tracking changes so just exported all .py scripts in exp into one file, then ran black code formatter on it and imported that file (nb_formatted.py).
ie for lesson 12 notebooks import exp.nb_formatted
Workflow:
Work through notebooks
If make any changes:
Export as .py to dl2/scripts/
Run dl2/scripts/clean_nb_output.py to remove irrelevant ipynb code
Step to relevant source in .py file using keybaord shortcuts
Optionally debug .py file with breakpoint if difficult to follow
NB debug only works pre .to_device, once on GPU not on CPU to be able to set breakpoints
dl2/
.ipynb's with minor changes (added comments and fixes to improve readability for me)
/utils
Python files containing functions and classes all in one place
Ideally would organise these files better but hacked it to get get through in reasonable time
/scripts
exported .ipynb's as .py to use in ide for debugging/stepping to source in ide
/exp