3 ms·
Image-based development drives me up the wall. On some level I love Jupyter notebooks, on another level I hate them. Since users can go up and down and back
by PaulHoule 1y ago
Image-based development drives me up the wall.
On some level I love Jupyter notebooks, on another level I hate them. Since users can go up and down and back and forth they frequently get the notebook into a state which is not reproducible. One answer is to frequently refresh the notebook from the beginning and re-run it, which might not work for the data scientist who is training a neural network that might take 12 hours to train. Another answer is that it should work like an Excel notebook, that is you change one thing and it should change everything that depends on that.
Practically I've struggled with issues like "they want to check in the notebook into version control with all the data and results in it so they can look at the notebook in Github and see a complete analysis" with "being able to check in changes to a notebook without an explosion of conflicts".
It can be so liberating to just start from zero in any situation and know it is all clean.