iPython

Here is the truth about IPython: the best place to learn Python is not a file – it’s a prompt that remembers what you just did, shows you plots, and lets you edit the last thought. IPython is that prompt, and Jupyter is the notebook built around it.

What it is, in plain words

IPython (Interactive Python) is an enhanced shell for Python, started by Fernando Pérez in 2001. Type code, see results immediately, use tab-completion, inspect objects with ?, run shell commands with !, and keep a searchable history. Jupyter Notebook and JupyterLab grew out of IPython’s notebook interface and now support many languages; the Python kernel is still IPython.

In 2026 you mostly meet IPython in three forms:

the terminal app, ipython;
the kernel inside JupyterLab / Notebook / VS Code / Emacs;
embedded in tools like Spyder and some scientific IDEs.

Why it matters

Scientific Python – NumPy, SciPy, pandas, Matplotlib – was shaped around IPython’s workflow: try an idea, plot it, refine. Even if you write production code in modules, exploration happens here. Learning IPython’s tricks cuts the friction between “I wonder” and “I see”.

An everyday analogy: a lab bench with a whiteboard

A .py file is a finished lab report. IPython is the bench: reagents out, whiteboard beside you, every measurement scribbled down and revisable. When the experiment works, you copy the clean parts into the report.

flowchart LR
  T[type code] --> R[result + plot]
  R --> H[history / %recall]
  H --> T
  T --> N[export to .py / notebook]

A hands-on example

pip install ipython matplotlib numpy
ipython

In [1]: import numpy as np
In [2]: x = np.linspace(0, 2*np.pi, 100)
In [3]: x[:5]                          # quick peek
In [4]: def f(t):
   ...:     return np.sin(t) * np.exp(-t/5)
In [5]: f(x)[:5]
In [6]: f?                             # docstring / source
In [7]: %timeit f(x)                   # micro-benchmark
In [8]: !head -3 /etc/os-release       # shell escape
In [9]: %history -n 1-5                # what did I just do?
In [10]: %save experiment.py 1-5       # write it out

Useful magics (the % commands):

Magic	Purpose
`%time` / `%timeit`	measure speed
`%run script.py`	run a file in this session
`%debug`	drop into debugger after an exception
`%matplotlib inline`	(notebook) show plots in place
`%autoreload 2`	reload imported modules as they change
`%pwd`, `%cd`, `%ls`	navigate without leaving

In a notebook, start with:

%load_ext autoreload
%autoreload 2
import matplotlib.pyplot as plt

Then edit your module in an editor, re-run one cell, and the imports refresh – REPL-driven development, Python-style.

Magics versus Python

Anything starting with % is an IPython magic, not Python syntax. Line magics (%timeit expr) take the rest of the line; cell magics (%%timeit) take the rest of the cell in a notebook. When you paste code into a .py file, strip the magics or keep them only in notebooks.

The configuration lives in ~/.ipython/profile_default/ipython_config.py. A few favourite tweaks:

c = get_config()
c.InteractiveShell.confirm_exit = False
c.TerminalInteractiveShell.editing_mode = "vi"   # or emacs
c.InteractiveShellApp.exec_lines = ["%load_ext autoreload", "%autoreload 2"]

Beyond the terminal: notebooks and editor kernels

JupyterLab gives you cells, Markdown narrative and plots in one document – ideal for notes and teaching. VS Code and Emacs can attach to the same IPython kernel, so you edit a .py file with proper tooling and still “Send to REPL”. That hybrid – file as source of truth, IPython as playground – is the workflow I recommend once scripts grow past a few cells.

Sharing: .ipynb files are JSON and work with Git, but diffs are noisy. Tools like nbdime help; or keep the notebook for exploration and promote finished code into modules, importing them back with %autoreload.

Common misconceptions

“IPython and Jupyter are the same.” IPython is the Python kernel and terminal shell; Jupyter is the multi-language notebook front end.
“Notebooks are for production.” They’re for exploration and communication. Package real code into modules and test it.
"? is only for docs." obj?? shows source; *pattern*? searches names.
“The terminal IPython can’t plot.” Use %matplotlib with an interactive backend, or write figures to files.
“It’s slow.” The shell is not the bottleneck; your algorithms are. %timeit proves it.

Connections

Python, NumPy, SciPy.
Open Data Science, Machine Learning.
Emacs – run IPython in M-x run-python or via ein/jupyter.