Using Python on COSMA
There are a number of different Python versions available on COSMA. You can see these using the module available python
and module available anaconda
commands. To select one of these, you can module load python/VERSION
. At this point, running the python
command (or in some cases python3
) will give you your selected Python interpreter.
If you need a newer version of Python than is available, please ask cosma-support.
Python libraries
The installed versions of Python generally have a limited number of libraries installed. Therefore, if you have a requirement for additional libraries, you will need to install them yourself in a virtual environment.
Installing Python libraries
To install python libraries, the following steps can be used:
module load the Python you wish to use (e.g.
module load python/3.12.4
).Create a virtual environment:
python -m venv /path/to/myenv
. /cosma/apps/PROJECT/USERNAME is a good place for this.Activate the virtual environment:
source /path/to/myenv/bin/activate
2.1 If you use csh or tcsh, you’ll need to source activate.csh instead.Install the packages you require:
pip install mypackage
Run Python:
python
If you log in with a different session and want to use this environment, you will need to remember to activate it: source /path/to/myenv/bin/activate
Once you have finished, you can deactivate your environment: deactivate
If you wish to remove it permanently, you can rm -r /path/to/myenv
You can have multiple virtual environments. This allows you to install specific library versions according to your requirements.
Pre-compiled python packages (wheels)
There are some pre-compiled python packages (wheels) available for several modules which depend on C libraries and are not totally straightforward to install with pip. This includes mpi4py, h5py with MPI support and nbodykit.
These are located in /cosma/local/python-wheels
.
For example, to install mpi4py in a python/3.12.4 virtual environment, you can (assuming you have followed the above instructions and activated your venv):
module load gnu_comp/14.1.0 openmpi
pip install /cosma/local/python-wheels/3.12.4/openmpi-5.0.3-hdf5-1.12.3/mpi4py-3.1.6-cp312-cp312-linux_x86_64.whl
If you also need h5py with MPI support then you can do
pip install /cosma/local/python-wheels/3.12.4/openmpi-5.0.3-hdf5-1.12.3/h5py-3.11.0-cp312-cp312-linux_x86_64.whl
Compiled modules
If the library you are pip installing requires compiling, you may need to load a C compiler module, and possibly various othe libraries (e.g. MPI, FFTW, HDF5, etc).
It is usually best to use the compiler that the Python interpreter you are using was compiled with. To find that out, start the Python interpreter, and the information lines will tell you which compiler to use.
Other compilers may offer improved performance, and you can experiement with these.
h5py installation
Building h5py can be done using the existing hdf5 modules. For example:
module load python/3.12.4
python3 -m venv h5py_test
source ./h5py_test/bin/activate
module load gnu_comp/11.1.0 hdf5/1.12.0
pip cache purge
pip install --no-binary=h5py h5py