Skip to content

Getting Started with MiniForge at REPACSS

Introduction

This guide covers setting up and using MiniForge (a minimal conda distribution) for data science applications on REPACSS.

This covers user-installed Python-based data science, machine learning, and scientific computing packages - requires local installation. For system-provided HPC applications, see Module System.

Why Use MiniForge

MiniForge offers several advantages over traditional Anaconda distributions:

  • Community-Centric: Uses conda-forge as the default source of packages, providing access to the latest community-maintained packages.
  • No Default Channel Conflicts: Avoids Anaconda's proprietary defaults channel, eliminating potential licensing issues.
  • Minimal Footprint: Lets you install only what you need, keeping your environment clean and efficient.
  • Fast and Reliable: Includes Mamba (C++ implementation) for dramatically faster package resolution compared to standard conda.
  • Licensing Clarity: No commercial restrictions or licensing concerns for research and commercial use.

🚀 Quick Start

If you just want to get started quickly:

  1. Install MiniForge: See Installing MiniForge section
  2. Create an environment: conda create -n myenv python=3.11
  3. Activate it: conda activate myenv
  4. Install packages: conda install numpy scipy matplotlib
  5. Start coding: python -c "import numpy; print('Success!')"

Overview of Conda Ecosystem

Conda Distributions and Channels

Figure source: Bilibili Video - Conda Ecosystem Overview

The conda ecosystem consists of different distributions and package channels that work together to provide package management capabilities:

Conda Distributions

  • Conda: The core cross-platform command-line tool for managing packages and environments, used by all distributions.
  • Anaconda Distribution: A commercial distribution that includes Conda and many preinstalled data science packages (NumPy, SciPy, etc.). By default, it connects to Anaconda's proprietary "defaults" channel.
  • Miniconda: A minimal Anaconda installer with basic tools. By default, it connects to Anaconda's "defaults" channel but can be configured to use other channels like conda-forge.
  • MiniForge: A minimal installer with only conda-forge preconfigured and no proprietary channels. Includes Mamba as a drop-in replacement for conda - a C++ implementation that dramatically increases package resolution speed compared to the standard conda solver.

Package Channels

  • conda-forge: A community-maintained channel providing the latest packages with open licensing. Used as the default by MiniForge and recommended for Miniconda.
  • Anaconda defaults: The default commercial channel for Anaconda and Miniconda, including "main", "R", and "msys2" sub-channels.

Removing Previous Conda Installations

If you previously installed Anaconda or Miniconda, we recommend removing them to avoid conflicts.

Check for existing Conda

conda --version

Locate old installations

ls -al | grep conda

Remove directories

find . -maxdepth 1 -name '*conda*' -exec rm -ir {} +
find . -maxdepth 1 -name 'Miniforge*' -exec rm -ir {} +
find . -maxdepth 1 -name 'Mambaforge*' -exec rm -ir {} +

Clean .bashrc

Remove any conda initialization lines and sign out and back in.

Installing MiniForge

Download and run installer

wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh
bash Miniforge3-$(uname)-$(uname -m).sh

Accept the license agreement and install to the default directory (e.g., $HOME/miniforge3).

Initialize and prevent auto-activation

source ~/.bashrc
conda config --set auto_activate_base false

Confirm installation

which conda
conda --version

Adding Channels (Optional)

By default, MiniForge uses conda-forge. Additional channels can be added if needed.

Set strict channel priority

conda config --set channel_priority strict

Optional: Add Bioconda (for bioinformatics)

conda config --add channels bioconda

Creating Virtual Environments

Create a new environment

conda create -n <env_name> python=3.11 numpy scipy matplotlib

Activate the environment

conda activate <env_name>

List environments

conda env list

Managing and Removing Environments

Deactivate the current environment

conda deactivate

Remove an environment

conda env remove --name <env_name>

📦 Installing Packages with Pip

When a package is unavailable via conda, you can use pip within an activated conda environment:

# Activate your conda environment first
conda activate myenv

# Install packages with pip
pip install somepackage

# Install from requirements.txt
pip install -r requirements.txt

Important

Always use pip inside an activated conda environment to prevent conflicts with system Python.

Optimizing Conda Initialization

Avoid installing all packages in the base environment. Create small, purpose-specific environments for each project.

Prevent base from activating on login

conda config --set auto_activate_base false

Check the configuration

conda config --show | grep auto_activate_base

Best Practices

  • Keep the base environment minimal
  • Create separate environments for different projects
  • Use descriptive environment names
  • Document your environment dependencies

Notes on Anaconda Licensing

Anaconda, Inc. enforces licensing terms on their default channels and tools. Usage in commercial or research contexts may require a paid license.

To avoid licensing complications and ensure access to current packages, we recommend avoiding the Anaconda defaults and using MiniForge + conda-forge.

Warning

REPACSS does not cover licensing costs related to Anaconda. Users are responsible for any commercial usage compliance.

For more details, see: Anaconda Terms of Service