Miniforge

Miniforge is the preferred installation method for Mamba and conda-forge and includes conda and mamba, along with their dependencies. It contains a number of useful packages which make it popular in fields like data science, machine learning and scientific computing.

Anaconda and Miniconda are no longer available on Apocrita due to licensing issues.

Miniforge is available as a module on Apocrita.

Usage

Loading the module

Python distribution module file conflicts

To prevent errors when running the Python interpreter, attempting to load an additional Python distribution module after loading a Miniforge module will produce a module load conflict error.

To run the default installed version of Miniforge, simply load the miniforge module:

module load miniforge

You can then check your python version:

$ which python
/share/apps/rocky9/general/apps/miniforge/24.7.1/bin/python
$ python -V
Python 3.12.6

$OMP_NUM_THREADS

The variable OMP_NUM_THREADS is set to the value of NSLOTS in serial jobs if the variable does not already exist in the environment. You will see a confirmation message when the module is loaded.

Conda package manager

In addition to the Conda package manager, Miniforge provides access to pip.

Mixing conda/mamba install and pip install

When using Conda, it is best to stick to only using conda install or mamba install for all packages. Try not to mix conda/mamba install and pip install if possible.

If you have no choice but to mix the two, please create your Conda environment specifying a version of Python, e.g.:

mamba create --quiet --yes --name myenv python=3.11.0

This will ensure that your environment has Python correctly setup and any pip install commands will write those packages inside your activated Conda environment. Failure to do this means they may end up in ${HOME}/.local/lib/<python version> which is highly likely to irreparably break all personal Python environments, as well as stop Jupyter Open OnDemand sessions from being able to load.

Environments

When working with Miniforge you'll probably want to update installed packages as well as installing some new ones. Since Miniforge is installed to shared storage and you don't have write access there, you'll need to create Conda environments somewhere where you do have write access (for example scratch directories or lab shares). You can use your home directory for this but beware: you have limited storage there.

Python virtualenv environments can also be created using Miniforge because it also provides the Python programming language as well as Conda environments.

Listing environments

You can list existing Conda environments as follows:

$ mamba env list
# conda environments:
#
base                     /share/apps/rocky9/general/apps/miniforge/24.7.1

Let's set up a new location for environments; we do this by editing .condarc which is in your home directory:

Do not use the defaults channel

Previous documentation may have led you to use the defaults channel in your ~/.condarc file. This is actively recommended against in the official Mamba documentation. Instead, you should use nodefaults, which will disable the defaults channel and use only the conda-forge channel.

$ cat ~/.condarc
channels:
  - nodefaults
ssl_verify: true
envs_dirs:
  - /data/scratch/abc123/anaconda/envs
pkgs_dirs:
  - /data/scratch/abc123/anaconda/pkgs

Here we have specified /data/scratch/abc123/anaconda as a location to install to.

Creating a new environment

Do not use a login node for creating Conda environments

Creating a Conda environment requires a reasonably large amount of time and memory. Do not use a login node for creating a Conda environment. Please use an interactive qlogin session.

Environment creation is single-core, so 1 core for 24 hours (1 hour is often not enough for more complex environments) with 8GB RAM is recommended:

qlogin -l h_vmem=8G -l h_rt=24:0:0

Let's create a new environment:

$ mamba create --quiet --yes --name myenv
Preparing transaction: ...working... done
Verifying transaction: ...working... done
Executing transaction: ...working... done

Here we have specified that we want a new environment called "myenv". We've not requested any packages. Suppose we wanted a specific point release of python installed. We could do:

$ mamba create --quiet --yes --name myenv python=3.11.0
Preparing transaction: ...working... done
Verifying transaction: ...working... done
Executing transaction: ...working... done

Let's see which version of python we have in our $PATH:

$ which python
/share/apps/rocky9/general/apps/miniforge/24.7.1/bin/python
$ python -V
Python 3.12.6

Let's activate the new environment. The asterisk denotes the currently active environment:

$ mamba activate myenv
$ mamba env list
# conda environments:
#
myenv                 *  /data/scratch/abc123/anaconda/envs/myenv
base                     /share/apps/rocky9/general/apps/miniforge/24.7.1

And let's check python again:

$ which python
/data/scratch/abc123/anaconda/envs/myenv/bin/python
$ python -V
Python 3.11.0

The following command will remove an unwanted environment, to save disk space:

$ mamba env remove -n myenv

Remove all packages in environment /data/scratch/abc123/anaconda/envs/myenv:

Everything found within the environment (/data/scratch/abc123/anaconda/envs/myenv),
including any conda environment configurations and any non-conda files, will be
deleted. Do you wish to continue?
(y/[n])? y

Installing packages using mamba

mamba and conda

The examples below use mamba, which is an improved solver, but conda will also still work (and has integrated the mamba solver since version 23.10.0). We still recommend users use mamba where possible, it should be a drop-in replacement for all conda commands.

You can install other packages from Conda into your environment.

Packages must be installed into activated environments

Packages must be installed into your personal environments. If package installs are attempted without first activating an environment, a permission error will be shown.

This example demonstrates installation of the scipy package:

$ mamba activate myenv
(myenv) $ mamba install --quiet --yes scipy
Preparing transaction: ...working... done
Verifying transaction: ...working... done
Executing transaction: ...working... done

(myenv)$ python -c "import scipy; print(scipy.__version__)"
1.14.1

The mamba list command will show all of the packages installed into your environment.

Example jobs

Threading with OpenMP

Some Python packages use OpenMP for threading. For serial jobs, the module sets the variable OMP_NUM_THREADS to be the number of requested slots in the current job, if the variable is unset when loading the module. This avoids an issue where some packages incorrectly use too many threads for OpenMP work.

If you are using OpenMP with such packages or in your own code, you should check that the OMP_NUM_THREADS variable has been set correctly, or override this value manually, either before or after loading the module.

Serial job

Here is an example job running on 1 core:

#!/bin/bash
#$ -cwd
#$ -j y
#$ -pe smp 1
#$ -l h_rt=1:0:0
#$ -l h_vmem=1G

module load miniforge
mamba activate myenv
python example.py

Serial job demonstrating environment switching

This example shows that you can switch Conda environments in a job script:

#!/bin/bash
#$ -cwd
#$ -j y
#$ -pe smp 1
#$ -l h_rt=1:0:0
#$ -l h_vmem=1G

module load miniforge
mamba env list
mamba activate myenv
mamba env list

If we view the job output:

# conda environments:
#
myenv                    /data/scratch/abc123/anaconda/envs/myenv
base                  *  /share/apps/rocky9/general/apps/miniforge/24.7.1

# conda environments:
#
myenv                 *  /data/scratch/abc123/anaconda/envs/myenv
base                     /share/apps/rocky9/general/apps/miniforge/24.7.1

Here we can see from the output of the mamba env list commands, that we are successfully switching environments inside a job script.

References

• Miniforge GitHub repository
• Mamba website
• conda-forge website