This post is inspired by a video from 2017 PyData conference in Berlin. Here I focus on several main points.
<aside> ☝ How big should a notebook file be?
Hypothesis — Data — Interpretation
</aside>
<aside> ☝ Keep your notebooks small!
(4-10 cells each)
How?
</aside>
I found this part particularly surprising, because my previous notebooks accompanying research papers have been huge. But by looking into his talk, I accepted this viewpoint.
Example: a fat notebook is split into several files in one directory.
Cache and images are separate folders.
<aside> ☝ Use shared libraries.
</aside>
Typical structure of the
ipynbfile.