There are a number of python versions installed on Lengau compiled with either gcc or the intel compiler, including Python 2.7.11 to 2.7.15, Python 3.5-3.7 and anaconda installs for Python 2 and 3. Use this command to get a full list:

  module avail chpc/python

All python modules should have mpi4py, numpy and scipy installed; however if matplotlib is required, please use anaconda (which also has the biggest range of 3rd-party python modules). The module chpc/python/3.5.2_gcc-6.2.0 also has a working matplotlib installation.

To use tensorflow, the anaconda module for python3 is also recommended.

The following will not work with the anaconda modules, see below if you want to use those.

Because CHPC users don't have permission to modify the python installations, you will not be able to install libraries to the python installation. You can, however, install them to a different location using the –user flag when you run pip install or pip3 install(the latter for python3):

pip install --user PySomeProject

This can only be done from a node with internet access - for example chpcviz1 or chpclic1 (but not a compute node). Of course, you will also need to replace “PySomeProject” with the name of the package that you want to install.

By default this installs to the .local directory within your home directory. This can be a problem if you need to install a lot of python packages, as your home directory can only accommodate 15GB. If you think you are likely to run out of space, you should rather specify a directory on lustre, as follows:

pip install --install-option="--prefix=/mnt/lustre/users/USERNAME/.local" PySomeProject

As before, replace USERNAME and PySomeProject with appropriate values. You can find more information on user installs in the official python documentation and more information on pip user installs here.

As an alternative, you can also set up a python virtual environment. This essentially gives you your own python installation which you can alter as you see fit. Again, it is perhaps best done in your lustre directory if a large number of python libraries will be installed.

To create a virtual environment called myenv, just cd to the place where you want the virtual environment to live (eg. cd /mnt/luster/users/USERNAME), and run the following command:

python3 -m venv myenv

You only need to do this once, to set up the virtual environment.

Then, to activate the virtual environment (which you will need to do whenever you use it - so put it in your submit-script or ~/.bashrc file):

source /mnt/lustre/users/USERNAME/myenv/bin/activate

Again, there is more information in the official python documentation.

Modules for anaconda version 2 and 3 are installed on lengau.To load these modules, use the following command (either from the command-line in an interactive session, or in a script:

  module add chpc/python/anaconda/2

or:

  module add chpc/python/anaconda/3-2019.10

Using virtualenv with anaconda is not recommended. However, Anaconda does include a similar functionality, which allows users to create an anaconda virtual environment.

The following instructions to set up your virtual environment.

YOU NEED ONLY DO THEM ONCE. Please replace USERNAME with your actual username wherever it occurs below.

1. Log into lengau.chpc.ac.za

2. ssh chpclic1 Unlike the compute nodes, chpcviz1 and chpclic1 have a connection to the internet, which is needed here.

3. Load the appropriate anaconda module

  module load chpc/python/anaconda/3-2019.10

You can substitute chpc/anaconda/2 if you want Python 2.

4. Do the once-off initialisation for your shell:

  conda init bash

You might have to log out and back in at this point to enable the shell setup.

5. Create a new conda environment:

  conda create --prefix /home/USERNAME/myenv python=3.6 anaconda

Change python=3.6 to python=2.7 if you're using anaconda/2.

6. Activate the environment

  conda activate /home/USERNAME/myenv

You don't have to use “myenv”, just replace it consistently with whatever name you prefer for your virtual environment, myenv will work, though.

7. You can now install whatever packages you need, that are not installed already; for instance, for rpy2, type:

 conda install rpy2

8. you can now type:

conda deactivate /home/USERNAME/myenv

to exit the virtual environment.

Now, once you've done this, you can log out, and ready your job for submission from the login node.

You must do the following in your job-submit script or your .bashrc file in your home directory:

Put it all near the top of your script, but after the #PBS directives:

  module add chpc/python/anaconda/3-2019.10

  conda activate /home/USERNAME/myenv

and at the end:

  conda deactivate /home/USERNAME/myenv

Then submit your script as normal with qsub.

if you ever need to install new Python packages, just repeat steps 1, 2, 3, 6, 7 and 8 from above. (omit 4 and 5).

There is known bug inflicting some versions of conda. Use

source /apps/chpc/chem/anaconda3-2019.10/etc/profile.d/conda.sh 

instead of conda init to initialise conda.

Replace the above path /apps/chpc/chem/anaconda3-2019.10 with the path to the Anaconda version you are actually using.

Remember that you can find the paths used by any module using the module show command:

  module show chpc/python/anaconda/3-2019.10