autoreload

Making module development less painful

Today’s blog post will be a bit shorter than usual. The main character of today’s post are two lines of Jupyter notebook magic.

TL;DR

The autoreload extension automatically re-imports a Python module if it changed during the lifetime of the notebook’s kernel. This is really handy if you are checking the module that you are developing with a notebook. Add these two lines on top of every notebook and be done with it.

%load_ext autoreload
%autoreload 2

The full story

But let’s start at the beginning: what are Jupyter notebooks? It is a bit difficult to get around them nowadays when working with Python, but not everyone comes from a Python background. Jupyter notebooks are a great way to combine code, text and results; especially for pedagogic purposes. They work a bit like a Mathematica notebook (well, they are both called notebooks after all). Most of the tutorials in Python package documentations are rendered versions of notebooks. See here for example (You can think about qiskit whatever you want. The documentation is quite nice.)

Coming back to our main character, the autoreload extension. By adding the following two lines at the very beginning of a Jupyter notebook, all Python modules are reloaded before a Python cell is executed:

%load_ext autoreload
%autoreload 2

In general, that does not sound super useful since numpy and scipy do not change on short notice. If you are developing your own code, however, it makes a huge difference. Let’s imagine you have one cell that imports your script foo_file and you would like to test the function bar.

from foo_file import bar

After you imported a module, it will not be imported again. Thus, the kernel of your notebook will not execute the import again. Usually that is fine, since your module does not change. You can execute your test

assert bar(3,9)==18

and everyone goes home happy.

If you are bug fixing foo_file and change the function bar, the situation looks a bit different. You will have to restart the kernel of the notebook to import bar again. The two lines with autoreload solve that problem for you. Now, you can just execute the cell with bar again and the most recent code is executed. That saves a lot of kernel restarting, and more importantly, headaches because bugs that are supposedly fixed do not disappear.

Patrick Emonts
Patrick Emonts
Postdoctoral Researcher

My research interests include tensor networks, lattice gauge theories and quantum information.