Installing conda

conda-forge is a “channel” for the conda package manager containing the Scientific Python ecosystem, Scikit-HEP, and even ROOT with carefully aligned package versions to ensure that you get a consistent, working system. Within a conda environment, you can still use pip to install packages that are not in this channel, thereby getting access to everything in the Python Package Index (PyPI), and everything in the conda environment is kept isolated from all other Python environments, so that you don’t disturb any applications that rely on a version of Python that ships with your operating system.

The software in conda-forge are not subject to Anaconda’s licensing restrictions, and the conda package manager is free software, so both can be used without any legal restrictions in national labs and universities.

Until recently, the (relatively) hard part had been to ensure that you’re using conda-forge, rather than an Anaconda default channel. The instructions below describe how to install Miniforge, which is conda-forge without the Anaconda default channel.

You likely have a package manager for your operating system, such as Homebrew, apt-get, or yum. Use conda for your Python packages and your operating system’s package manager for applications (web browsers, text editors, etc.).*

(* We’re doing conda a disservice by describing conda as a Python package manager, though it does much more, for the sake of keeping this description simple.)

We recommend using mamba, which is a drop-in replacement for conda that is many times faster (in the “Solving environment: …” step). You particularly notice it when a package has many dependencies or complex version constraints on its dependencies.

In fact, the conda developers are incorporating mamba into conda. At the time of this writing, however, that integration is still experimental. These instructions will describe how to use mamba directly.

The entire Python distribution, with all packages and the binary shared libraries that support them, will go into a new directory, most likely in your home directory and named mambaforge. All of the files in it are installed with your own user permissions (i.e. not superuser/requiring sudo).

1. Delete that directory with rm -rf ~/mambaforge.
2. Delete a file named ~/.condarc, if you have one.
3. Check your shell configuration file, probably named ~/.bashrc, for a “>>> conda initialize” section. If you have one, delete it.

Those three steps will remove any vestige of the conda installation.

If you already have a conda installation, you can bundle your current environment into an environment file (a list of names and versions of packages) with

conda env export --from-history > old-environment.yml

After setting up a new conda system, you can reinstall all of those packages/versions with

conda env create -f old-environment.yml

If you say “yes” to let the installer script modify your shell configuration, then the next terminal you open will be in the conda environment. For instance,

python

will run the conda environment’s Python, rather than any other Python you have installed on your computer. This is what conda calls the “base” environment (though you can create more environments that are independent of this one).

If, instead, you want to explicitly opt-into conda environments by calling a command, use

conda config --set auto_activate_base false

to prevent the “base” environment from being automatically loaded in each new terminal. Now all environments, including “base”, have to be explicitly activated with

conda activate name-of-environment

See managing environments in the conda documentation for more.

If you say “no” to not let the installer script modify your shell configuration, then you will have to manually find the path to the conda executable, which is in ~/mambaforge/bin/conda. All of the above applies, but your shell might not be able to find conda or python.