Micromamba

Micromamba is a statically linked C++ executable that provides the full power of the Conda ecosystem without needing a pre-installed Python "base" environment. Unlike pip, uv and other package managers, Conda can manage libraries written in C, C++, or R, which is vital for machine learning and AI work.

INFO

When to use: Micromamba is preferred for projects requiring polyglot or non-Python dependencies; for repositories strictly focused on the Python ecosystem, uv remains the optimized choice.

Installation

Arch Linux

yay micromamba-bin

Initialize and Verify

micromamba shell init -s bash

INFO

If the command above fails, run this before repeating the same step again: export MAMBA_ROOT_PREFIX=~/micromamba

source ~/.bashrc
micromamba --version

If you need activation support only for current shell session, you can load shell hook manually instead:

eval "$(micromamba shell hook -s bash)"
micromamba activate <env_name>

Channels

Micromamba pulls packages from channels. In practice, many scientific and data packages are commonly installed from conda-forge.

micromamba config append channels conda-forge
micromamba config set channel_priority strict

TIP

Strict channel priority helps keep dependency resolution more predictable.

Computational Workstations & Shared Environments

On shared computational workstations (e.g. kmer), micromamba is often pre-installed with a shared conda root.

Verify installation

micromamba --version
micromamba info

Make sure the root prefix is set correctly (example):

/home/ndemo/.conda

Create a personal environment

micromamba create -n test python=3.12 numpy scipy -y

Activate and verify it:

micromamba activate test
which python
python --version

The expected path should look like:

/home/ndemo/.conda/envs/test/bin/python

Deactivate when done:

micromamba deactivate

Create and use shared environments (managed by root)

On shared servers, system administrators can create environments in a shared location (e.g. /data/conda/shared_envs/) that all users can activate.

WARNING

Only root/administrators can create or update shared environments.

As root:

micromamba create \
  -p /data/conda/shared_envs/scientific \
  -c conda-forge \
  python=3.12 numpy scipy pandas matplotlib \
  -y

As any user, you can activate it directly by path:

micromamba activate /data/conda/shared_envs/scientific

To add packages to a shared environment, only the administrator can do so:

micromamba install \
  -p /data/conda/shared_envs/scientific \
  scikit-learn -y

Core Workflow

WARNING

Don't "brainlessly" run the following commands, since you need to replace the <> values.

Environment Management

Create

Standard

# The basics: environment creation
micromamba create --name <env_name> python=3.11

Create and install packages

# You can target specific packages, versions and channels (like conda-forge)
micromamba create -n <env_name> -c conda-forge python=3.11 <packages>

Build from YAML

# Environments can be built given an existing .yml file
# NOTE: You can specify the environment name adding "-n <env_name>".
micromamba create -f <filename>.yml

Local

# Keep the environment physically inside the project folder 
# (like a .venv) instead of Micromamba's central registry.
# NOTE: To activate this later, you will need to point to the path: 
#   micromamba activate ./env
micromamba create -p ./env python=3.11 <packages>

Example environment.yml:

name: myenv
channels:
	- conda-forge
dependencies:
	- python=3.11
	- numpy
	- pip

Once created, don't forget to always activate your environment:

micromamba activate <env_name>

TIP

Micromamba also allows you to run a command inside an environment without activating it first:

micromamba run -n <env_name> python <script_name>.py

At some point, you may also want to temporarily "quit" your environment:

micromamba deactivate

Delete

Standard

# Remove the environment. Cleanly
micromamba env remove -n <env_name>

By name

# You can target the environment either by its name or its directory path
micromamba env remove -p ./env_folder

TIP

Alternatively, you can skip the word env entirely by utilizing the standard package removal command paired with the --all flag (which instructs Micromamba to strip away all packages, wiping out the environment).

Extra

Environment List

# For the sake of clarity, better check the existing ones
micromamba env list

Export to YAML

# As they can be imported, environments can be also exported to .yml files
micromamba env export > environment.yml

Package Management

Install

# Install a package. Or more, just list them
# You can optionally specify their version (<package_name=version_number>)
micromamba install <package_name>

Remove

# Remove a package from the environment
micromamba remove <package_name>

Update

# Update a package. Better than reinstalling it
micromamba update <package_name>

Search

# Search for packages and channels
micromamba search <package_name>

List

# Gives a list of all the packages in the environment
micromamba list

Cache

# Installing packages builds up cache. Give it a flush, sometimes
micromamba clean --all