Python

Python is a popular and versatile programming language used in many scientific disciplines across the university. It is by far the most popular language in use by staff at Southampton and is the most used tool on Windows desktops too!

The university Linux desktop platform has a large number of Python implementations for you to choose from. We’ve worked hard to try to provide the options you might need and granted you the privileges needed to install packages as required.

Note

To install python modules via sudo pip you need to be in the Administrators group. See Permissions for more information. You don’t need to be in this group to use the conda command. You can also use pip directly, without sudo, and install packages in your home directory.

List of available implementations

  • CPython

    • RHEL system python - 2.7.5
    • Miniconda (Anaconda) - 2.7, 3.6
    • IUS packages - 3.4.6, 3.5.3, 3.6.0
    • SCL packages - 2.7.8, 3.3.2, 3.4.2, 3.5.1
    • EPEL package - 3.4.5

  • PyPy

    • PyPy 5.6 - 2.7.12
    • PyPy3 5.5-alpha - 3.3

  • Jython

    • SCL Jython 2.7

You can use Desktop Manager to install the various options, or read below for the yum commands to use instead.

Which implementation should I use?

  • If you want to use Python 2, and you don’t need conda packages, then use the RHEL system Python, and use pip to install packages.
  • If you want to use Python 3, and you don’t need conda pacakges, then use the IUS Python 3.6 package and use pip3.6m to install pacakges
  • If you need conda packages (Anaconda Python packages) then use miniconda

RHEL system Python

Unless you need Python 3 or conda packages then we recommend you use the system python in /usr/bin/python. This is supported by Red Hat and kept up to date with security and bug fixes.

You can install packages by running sudo pip eg:

sudo pip install numpy

IUS Python

The IUS repository provides Python 3 packages for Red Hat Enterprise Linux. If you need to use Python 3 then we recommend you use the IUS Python packages. Several versions are available for install:

  • Python 3.4 - sudo yum install python34u python34u-pip
  • Python 3.5 - sudo yum install python35u python35u-pip
  • Python 3.6 - sudo yum install python36u python36u-pip

You can then use Python via the following paths to the binary:

  • Python 3.4 - /usr/bin/python3.4
  • Python 3.5 - /usr/bin/python3.5
  • Python 3.6 - /usr/bin/python3.6

The version of Python in /usr/bin/python3 depends on the order the IUS packages are installed, so we don’t recommend you use that path.

You can install packages via sudo pip:

  • Python 3.4 - sudo pip3.4
  • Python 3.5 - sudo pip3.5
  • Python 3.6 - sudo pip3.6

For example, to install numpy on IUS Python 3.6:

sudo pip3.6 install numpy

PyPy

PyPy is a just-in-time (JIT) compiler for Python which aims to be compatible with the standard CPython implementation. Using PyPy often yields significant performance benefits.

To install PyPy run the following command:

sudo yum install pypy256

This installs PyPy 5.6 which is compatible with CPython 2.7.12.

To install PyPy3 run the following command:

sudo yum install pypy3355

This installs PyPy3 5.5-alpha which is compatible with CPython 3.3.5.

To run PyPy you can either use the following paths:

  • /usr/bin/pypy56 - PyPy 5.6 (Python 2.7.12)
  • /usr/bin/pypy3355 - PyPy 5.5-alpha (Python 3.3.5)

Or you can use the module system:

  • module load pypy/5.6 - PyPy 5.6 (Python 2.7.12)
  • module load pypy/3.3-5.5 - PyPy 5.5-alpha (Python 3.3.5)

Once you’ve loaded the module you can use the pypy command directly and it will correspond to the version of pypy you chose to load.

You can install packages with pip:

  • sudo /local/software/pypy/5.6/bin/pip - PyPy 5.6 (Python 2.7.12)
  • sudo /local/software/pypy/3.3-5.5/bin/pip3 - PyPy 5.5-alpha (Python 3.3.5)

For example, to install numpy on PyPy 5.6:

sudo /local/software/pypy/5.6/bin/pip install numpy

Miniconda (Anaconda)

Anaconda, and its conda package manager, is an alternative to the pip package manager (although the developers insist it complements pip and solves different problems).

Anaconda, conda and miniconda are not designed for “system-wide” use, they are intended for individual users to download and install for just that one user. Better multi user support is planned, but as of March 2017, is not yet available.

To use conda you don’t need to download and install Anaconda or miniconda yourself, instead you can get started by installing miniconda which just contains conda and python:

sudo yum install miniconda

Once installed you can use conda to create a new environment within your home directory with whatever version of Python and whichever Python packages you need.

You should start by creating an environment, e.g:

conda create -n yourenvname python=x.x

You specify the version of python you want with the python=x.x flag, but this is optional and if omitted conda will use the version of Python shipped with miniconda which at the time of writing is Python 3.6.0.

You can specify at creation time the packages you want installed. For example, to install the entire anaconda set of packages you can do the following:

conda create -n yourenvname python=3.6 anaconda

You can then use your new environment like so:

source activate yourenvname

And you can then stop using it like so:

source deactivate

You can install additional packages with conda:

conda install -n yourenvname [package]

And if you want to delete an environment do the following:

conda remove -n yourenvname -all

SCL Python

Note

The SCL python33 package conflicts with the IUS Python packages. You cannot have both versions installed at the same time. To install the SCL python33 package you must first remove the IUS Python packages. In any case we strongly recommend you use the IUS packages rather than SCL.

Red Hat provides several CPython packages as part of its “Software Collections” system. These packages are generally older than the IUS packages and are more difficult to use - they require the use of the scl command. At the time of writing the following versions are available:

  • Python 2.7.8 - sudo yum install python27 python27-python-pip
  • Python 3.3.2 - sudo yum install python33
  • Python 3.4.2 - sudo yum install rh-python34 rh-python34-python-pip
  • Python 3.5.1 - sudo yum install rh-python35 rh-python35-python-pip

Once installed you can’t use python until you use the scl command which is somewhat like the environment module system:

  • Python 2.7.8 - scl enable python27 bash
  • Python 3.3.2 - scl enable python33 bash
  • Python 3.4.2 - scl enable rh-python34 bash
  • Python 3.5.1 - scl enable rh-python34 bash

Once you’ve run the scl command then the python command will now be the version of Python you requested. The pip command will also be updated for the SCL python, but it won’t work unless you use a special sclsudo command we’ve created. So to install packages you should run:

sclsudo pip install numpy

EPEL Python

Note

The EPEL python 3.4 package conflicts with the IUS Python 3.4 package. You cannot have both versions installed at the same time. To install the EPEL python you must first remove the IUS Python 3.4 packages. In any case we strongly recommend you use the IUS packages rather than EPEL.

Another alternative Python 3 package is provided by :doc`epel`. We recommend that you use the IUS packages instead since the EPEL Python package is now quite out of date. If you do want to use it, install it like so:

sudo yum install python34 python34-pip

You can then use the package with the binary path:

/usr/bin/python3.4

and you can install packages with pip:

sudo /usr/bin/pip3.4 install numpy

Jython

Jython is an implementation of Python running on the Java virtual machine (JVM). Red Hat have provided a packaged version of Jython as part of their Developer Toolset 4 product. To install it run this command:

sudo yum install devtoolset-4 devtoolset-4-jython

Then run the scl command to enable it:

scl enable devtoolset-4 bash

You can then run jython directly:

jython