Python

This page documents the “anaconda” installation on Bessemer. This is the recommended way of using Python, and the best way to be able to configure custom sets of packages for your use.

“conda” a Python package manager, allows you to create “environments” which are sets of packages that you can modify. It does this by installing them in your home area. This page will guide you through loading conda and then creating and modifying environments so you can install and use whatever Python packages you need.

Using Conda Python

After connecting to Bessemer (see Establishing a SSH connection), start an interactive session with the srun --pty bash -i command.

Anaconda Python can be loaded with one of the following:

module load Anaconda3/2019.07
module load Anaconda3/5.3.0

The root conda environment (the default) provides Python 3 and no extra modules, it is automatically updated, and not recommended for general use, just as a base for your own environments.

Warning

Due to Anaconda being installed in a module you must use the source command instead of conda when activating or deactivating environments!

Creating a Conda Environment

Every user can create their own environments, and packages shared with the system-wide environments will not be reinstalled or copied to your file store, they will be symlinked, this reduces the space you need in your /home directory to install many different Python environments.

To create a clean environment with just Python 3.8 and numpy you can run:

conda create -n mynumpy python=3.8 numpy

This will download the latest release of Python 3.8 and numpy, and create an environment named mynumpy.

Any version of Python or list of packages can be provided:

conda create -n myscience python=3.5 numpy=1.15.2 scipy

If you wish to modify an existing environment, such as one of the anaconda installations, you can clone that environment:

conda create --clone myscience -n myexperiment

This will create an environment called myexperiment which has all the same conda packages as the myscience environment.

How to avoid large conda environments filling up your home directory

Home directories have limited space and can often reach their quota limit. Conda environments exponentionally take up space, if you have or want to create one or more large Conda environments (e.g. containing bulky Deep Learning packages such as TensorFlow or PyTorch) then there’s a risk you’ll quickly use up your home directory’s storage quota.

To avoid this, build your conda environments in a fastdata area

  1. Create a .condarc file in your home directory if it does not already exist.

  2. Add an envs_dirs: and pkgs_dirs: section to your .condarc file as shown below:

pkgs_dirs:
- /mnt/parscratch/users/$USER/anaconda/.pkg-cache/

envs_dirs:
- /mnt/parscratch/users/$USER/anaconda/.envs

  1. We recommend users create their own personal folder in the /fastdata area. As this doesn’t exist by default, you can create it with safe permissions. See fastdata area

  2. Then create .envs and .pkg-cache directories in your fastdata area as shown below:

mkdir -p /mnt/parscratch/users/$USER/anaconda/.pkg-cache/  /mnt/parscratch/users/$USER/anaconda/.envs

Installations of environments and package caching should now occur in your fastdata area

Installing Packages Inside a Conda Environment

Once you have created your own environment you can install additional packages or different versions of packages into it. There are two methods for doing this, conda and pip, if a package is available through conda it is strongly recommended that you use conda to install packages. You can search for packages using conda:

conda search pandas

then install the package using:

conda install pandas

if you are not in your environment you will get a permission denied error when trying to install packages, if this happens, create or activate an environment you own.

If a package is not available through conda you can search for and install it using pip, i.e.:

pip search colormath

pip install colormath

Using conda Environments

Once the conda module is loaded you have to load or create the desired conda environments. For the documentation on conda environments see the conda documentation.

You can load a conda environment with:

source activate myexperiment

where myexperiment is the name of the environment, and unload one with:

source deactivate

which will return you to the root environment.

It is possible to list all the available environments with:

conda env list

Provided system-wide are a set of anaconda environments, these will be installed with the anaconda version number in the environment name, and never modified. They will therefore provide a static base for derivative environments or for using directly.

Using Conda and Python in a batch job

Create a batch job submission script called myscript.slurm that is similar to the following:

#!/bin/bash
#SBATCH --ntasks=1
#SBATCH --time=10:00
#SBATCH --mem-per-cpu=100

export SLURM_EXPORT_ENV=ALL
module load Anaconda3/2019.07

# We assume that the conda environment 'myexperiment' has already been created
source activate myexperiment
srun python mywork.py

Then submit this to Slurm by running:

sbatch myscript.slurm

Further Conda Python Learning Resources

The resources and training courses below may be of interest: