Advice on how to work with Python

FluidDyn should be used also by scientists that are not experienced in Python. We provide some advice on how to work with Python and how to get a good Python environment.

Use an up-to-date Python environment!

Python is an old language but the strong dynamics in scientific Python is actually young. The base pakages have greatly improved these last years so it is really better to use recent versions. Therefore it is not a good idea to use scientific python libraries packaged in not very recent Linux versions.

Read!

Python 2.7 and Python 3

Python 3.6 is cleaner and better on many aspects that the 2.7 version of the language. The CPython interpreter for Python 3.6 is also faster for most of the tasks that the version for Python 2.7. Now (2017), for scientific purposes, I see no reason why not to use python >= 3.5. Conda is very convenient to use recent versions of Python.

Writing Python 2-3 compatible code is not difficult (with some try and from __future__ statements and using for example the package future) but yet it is time consuming and we cannot benefit from the nice new features of recent versions of Python. I tend to think that the time when the new versions of the fluiddyn packages drop compatibility with Python 2.7 is not far…

Use fly checks and a good editor!

Python is a very dynamics language. It is very nice but it is also very dangerous. A silly error (a misspell for example) is very easy and there is no compiler to tell you that something is wrong. Automatic checking of the code is enough to avoid most of these silly errors so anyone has to use it.

The style in Python is also really important (see below) so any Python developer has to get used to code properly. The best way is to code with a fly checker that tells you as soon as you do something wrong.

Most experienced Python programmers use an good Python editor with fly checking and it is really very useful. So of course beginners have to use a good Python editor running fly checks!

If you like integrated development environment, you can for example use Spyder (Scientific PYthon Development EnviRonment). Note that Spyder has to be setup correctly to use fly checks.

Another good solution is Emacs, but it should be setup correctly (for example with Flycheck, see my Emacs configuration).

Note that code checkers can also be used outside of the editor, for example with pylint. we can use the commands (from the FluidDyn root directory):

pylint fluiddyn
pylint -E fluiddyn.lab.probes
pylint --help-msg=no-member

The style is important

Most of the time, we have to follow the Style Guide for Python Code, the so-called “pep 8”). It is not just for fun. On the long term, it really helps.

In particular,

  • limit all lines to a maximum of 79 characters.
  • most of the time, comments before the code. At least not in very long lines of more than 79 characters.
  • names of the modules in lower case.
  • names of the classes in CamelCase, i.e. LikeThis.
  • no space before a comma and a space after.
  • no tabulation! four spaces.
  • no trailing white space.
  • documentation of the functions in a docstring.
  • Python is in English. It is a good idea to write Python modules all in English.

Use virtualenv (if you do not use conda)

It is a good idea to use a virtual environment. With virtualenv it is very easy. I do something like this:

MYPY=$HOME/path/mypy
mkdir -p $MYPY
virtualenv --system-site-packages $MYPY

Then I have a non-executable script load_mypy.sh in the directory ~/bin with something like this:

export VIRTUAL_ENV_DISABLE_PROMPT=0
MYPY=$HOME/path/mypy
source $MYPY/bin/activate

This script should be use by source ~/bin/load_mypy.sh so you can have a line with alias load_mypython=’source ~/bin/load_mypy.sh’ in the file ~/.bash_aliases.

Remark: it could also be convenient to use the module load ... procedure…

For Matlab users

If you begin with Python and you really like Matlab, I would advice to add in your .bashrc the line:

alias ipython='ipython --pylab'

With the option --pylab, Ipython imports the module matplolib.pylab with the command from matplotlib.pylab import * and runs matplotlib.pylab.ion() (this can take a few seconds), so the iterative Python console will behave much more like in Matlab than the standard ipython console without pylab imported.

In contrast, in you script, do not use the devil line from matplotlib.pylab import *. It is much better to learn how to use matplotlib with the import import matplotlib.pyplot as plt.