Jupyter and notebooks

A notebook is cells of code and text run one at a time, with the result left underneath. Perfect for exploring data: run a cell, look, change it, run it again.

The big trap: the execution order. The cells keep their state regardless of the order in which you run them. You can

• run cell 5 before cell 3,
• change cell 2 and not re-run it,
• delete the cell that created a variable you are still using.

The notebook looks right but cannot be re-run from the beginning. The numbers In [1], In [7], In [3] give it away.

Discipline that saves you:

1. Re-run everything from the beginning (Kernel → Restart & Run All) before you share it or draw conclusions. If it does not go through, the result is not reproducible.
2. Imports and constants in the first cell.
3. One cell = one step. Fifty lines in one cell is a script, not a notebook.
4. Move finished code out into a .py file and import it — the notebook gets short and the functions testable.

# the first cell
import numpy as np, pandas as pd
from minapp.data import las_och_rensa      # finished code lives in modules
DATA = "data/train.csv"

jupyter nbconvert --to script analys.ipynb   # notebook → .py
jupyter nbconvert --clear-output --inplace analys.ipynb   # clear the output before committing

Git and notebooks: the output (images, long tables) makes diffs unreadable and can contain data that should not go into the repo. Clear the output before committing, or use jupytext.