Create unit test / CI

[gimseng]

Perform CI (purest) for code reviews. I am not entirely sure about how to do that for Jupyter notebook (.ipynb) but was reading around the internet and it seems like catching the error when converting it to .py file is the way to go.

[gimseng]

Also pep8 lint test. Actually there's been a discussion on how to do that for Jupyter notebook. See https://stackoverflow.com/questions/26126853/verifying-pep8-in-ipython-notebook-code

[gimseng]

[x] Create flake8 (pep8) lint test whenever pull-request occurs. This applies to all .py file. See #30
[ ] Create flake8 for Jupyter notebook

[gimseng]

I think since we are not building a package or library, it doesn’t make send to do too much testing. At this point pep8 linting and if the codes run without error, that should be a good starting point.

[gimseng]

Finally found the solution to deal with module, testing and version control with notebooks. The nbdev package of fast.ai provides such a solution. See https://www.fast.ai/2019/12/02/nbdev/

It also generates documentations so that you can read the notebook like a HTML page with figs and etc. I guess that's how fast.ai courses are built.

We could potentially use this tool to build up a static markdown page as a textbook-like page.

[gimseng]

There are two other solutions that are closely related to the above development: https://github.com/jupyterlab/jupyterlab-git and https://www.reviewnb.com

Might be interesting to think about version control with this as well

[gimseng]

Finally, we can imagine going all colab. If notebooks can work out to be the main way to go and we use colab as our prefer 'IDE', nbdev + colab is a possible way to standardise everything.

See https://forums.fast.ai/t/using-nbdev-with-google-colab/61755/12