Jupyter Stack Cheat Sheet

1Install & launchLab 4 / Notebook 7

pip install jupyterlab # or notebook (Notebook 7)★
JupyterLab is the full IDE; Notebook 7 is the classic single-doc UI, now built on the same Lab components.
jupyter lab # or: jupyter notebook★
Starts the server + opens the browser. jupyter lab --no-browser --port 8888 for remote/servers.
jupyter lab list · jupyter --paths · jupyter kernelspec list
Running servers, config/data paths, and installed kernels.
jupyter labextension list
JupyterLab 4 extensions are pip-installable prebuilt packages (no Node build step).

2Notebook workflowcells & shortcuts

# Command mode (Esc) vs Edit mode (Enter)★
Two modes: Command (blue, act on cells) and Edit (green, type in a cell). Most shortcuts are command-mode.
Shift+Enter run · a/b insert above/below · dd delete · m/y md/code★
The essential command-mode keys. Ctrl+Shift+C opens the command palette for everything else.
Kernel → Restart & Run All★
The reproducibility check: run top-to-bottom in a fresh kernel to catch hidden out-of-order state before sharing.
display(obj) · last expression auto-displays · ; to suppress
The last line of a cell renders; end with ; to hide it (e.g. matplotlib return values).

3Magics & shell% and !

%timeit func() · %%time # time a line / whole cell★
% = line magic, %% = cell magic. %timeit auto-loops for a stable estimate.
%load_ext autoreload %autoreload 2★
Edit imported modules without restarting: auto-reloads changed source on each cell run — essential when developing a library alongside a notebook.
!pip install x · files = !ls *.csv★
! runs a shell command; capture its output into a Python list. Prefer %pip/%conda so installs hit the kernel's env.
%matplotlib inline · %run script.py · %debug · %who
Inline plots, run a file into the namespace, post-mortem debugger, list variables. %lsmagic lists all.

4Introspection & displayexplore objects

obj? # docstring + signature · obj?? # + source★
The fastest docs lookup. Tab completes; Shift+Tab shows the signature inline while typing a call.
_ · __ · _12 # last / 2nd-last / Out[12] results
Output history variables — reuse a previous cell's result without re-running.
from IPython.display import display, HTML, Image, Markdown display(HTML("<b>hi</b>"))★
Render rich output (HTML/Markdown/images/audio). _repr_html_ on your class controls how it displays.

6nbconvert (export)to other formats

jupyter nbconvert --to html notebook.ipynb★
Export to html, pdf (needs LaTeX), webpdf, markdown, script (.py), slides.
jupyter nbconvert --to notebook --execute nb.ipynb --output out.ipynb★
Run a notebook headlessly and save the executed copy — the basis of scheduled/CI notebook jobs.
--to script · --no-input # strip code cells (report view)
--no-input for a clean report; --to script to extract the code.

7nbclient (execute)programmatic runs

from nbclient import NotebookClient import nbformat nb = nbformat.read("in.ipynb", as_version=4) NotebookClient(nb).execute()★
The library that actually runs notebooks (nbconvert's --execute uses it). Programmatic control over kernel & timeouts.
NotebookClient(nb, timeout=600, kernel_name="python3").execute()
Set per-cell timeout and which kernel to run in. Pair with papermill to parameterize notebooks.

8nbformat (edit)notebooks as JSON

import nbformat nb = nbformat.read("nb.ipynb", as_version=4)★
A notebook is JSON; nbformat reads/validates it into an object with nb.cells, metadata, etc.
nb.cells[0].source · nbformat.v4.new_code_cell("print(1)")★
Inspect/build cells programmatically — generate notebooks or strip outputs before committing.
nbformat.write(nb, "out.ipynb") · nbformat.validate(nb)
Write it back / validate against the schema.

9ipytest & pipdeptreetest & deps

import ipytest; ipytest.autoconfig()★
Run pytest inside a notebook — great for teaching & quick TDD.
%%ipytest def test_add(): assert add(2,3) == 5★
The %%ipytest cell magic collects & runs the test functions defined in that cell.
pip install pipdeptree pipdeptree # tree of installed packages & their deps★
Visualize the dependency tree of the current env — find what pulls in a package.
pipdeptree --warn fail · pipdeptree -p somepkg --reverse★
Debug conflicts: --warn fail surfaces version clashes; --reverse shows what depends on a package (why it's installed).

!Common gotchasread before shipping

out-of-order execution
Cells run in whatever order you click. Hidden state hides bugs — Restart & Run All before trusting or sharing a notebook.
wrong kernel = "module not found"
The notebook's kernel is a specific Python env. Check sys.executable; register the env with ipykernel install.
%pip / %conda, not !pip
Shell !pip may install into the server's env, not the kernel's. The % magics always target the running kernel.
outputs bloat git diffs
Committed notebooks carry outputs & execution counts. Strip them (nbstripout / --ClearOutput) or diffs are unreadable.

Jupyter stack cheat sheet