1.6. Miniconda: Python(s) in a convenient setup

1.6.1. Overview

Python is increasingly used in the imaging world. Everyone and their sheepadoodle seems to be writing in it and coming across it in some part of their daily work. Part of its appeal is that there exist many useful modules for carrying out tasks (plotting, machine learning, statistical modeling, writing in XKCD font, etc.).

Along with the flexibility and broad appeal of Python comes a need for managing and maintaining dependencies: making sure the necessary dependencies are installed, adding/updating modules, and using the requisite version of Python. For example, there were major changes between Python 2 and 3 (just try evaluating 1/2 in each case), and occasionally across sub-versions code can break; we can only wait and see what will happen with Python 4.

Anyway, managing potentially multiple versions of Python, as well as diferent sets of dependencies can be tricky. There are different ways of doing this, but through personal experience I have found using “Conda” to be (by far) the easiest and most straightforward. We can basically set up one or more environments with specific Python versions, modules and dependencies, and we can swap back and forth between them fairly easily—they are each separate and distinct, and don’t get in each other’s way. We can set one to be “on” by default, not really think about it often, and then just change to another environment if we need to. Nice.

This page is neither a recommendation to use Conda, nor is it a full tutorial on using it. It is just meant to be a helpful starting point to Conda, if it sounds appealing to you, to manage your Python needs.

We will use the specific installation form of Conda called “Miniconda”, because it starts of being light-weight, and you can then add whatever you need.

1.6.2. Set up Conda (verbose)

Whether you are using Linux—either directly or through the Windows Subsystem Linux—or Mac, the setup will basically follow the same steps. (Miniconda can also be installed on Windows directly, but since AFNI doesn’t run there, who cares?)

This example is for setting up Conda for a single user system. Instructions for installing Conda for multiple users is described on the conda website here. We don’t discuss it further.

Note that you should not need your sudo password to install Conda on your system, nor to run later “conda” commands.

Download+install miniconda

  1. Click here and select installer to download (e.g., to home directory):

    Note on selection: Most computers now are 64-bit, and selecting the “Python 3.7” version makes sense (you can still use it to set up Python 2.* environments on your system). For Macs, I used the bash script version (rather than pkg), just because the former is more similar to what is done on Linux.

  2. Run the downloaded installer script:

    Type bash SCRIPT_NAME, such as:

    • For Linux:

      bash Miniconda3-latest-Linux-x86_64.sh

    • For Mac:

      bash Miniconda3-latest-MacOSX-x86_64.sh

    You will be prompted for various things:

    • hit Enter to read agreement; hit spacebar to navigate through it quickly; type yes to agree to it.
    • when prompted about installation directory, I just left it in the default “home”.
    • did I wish to initialize Miniconda3? I typed: yes

    I then got a message that I was successfully set up; and also a “conda: Command not found message”. So I did…

  3. Make updates known to terminal:

    Open a new terminal. You should see a text string like “(base)” to the left of your terminal prompt, and it will be to the left of every prompt (though you can optionally turn that off, which I do).

    You can also source the ~/.*rc file for your shell, which should update your current terminal. (To know what type of shell you are using, you can type echo $0.)

    • For bash:

      source ~/.bashrc

    • For tcsh:

      source ~/.cshrc

    You should see a string “(base)” string stuck before your terminal prompt now. Typing conda -V should also show you your version number.

    Note: your conda version should be at least 4.6.

Sidenote

What has Conda done to initialize things in the terminal? It has stuck some commands into your shell’s startup file; in my ~/.bashrc file (because I use bash shell), I can now see the following text:

# >>> conda initialize >>>
# !! Contents within this block are managed by 'conda init' !!
__conda_setup="$('/home/USERNAME/miniconda3/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
if [ $? -eq 0 ]; then
    eval "$__conda_setup"
else
    if [ -f "/home/USERNAME/miniconda3/etc/profile.d/conda.sh" ]; then
        . "/home/USERNAME/miniconda3/etc/profile.d/conda.sh"
    else
        export PATH="/home/USERNAME/miniconda3/bin:$PATH"
    fi
fi
unset __conda_setup
# <<< conda initialize <<<

… where USERNAME is replaced with my actual username. If you chose to install miniconda in a different location than your home directory, then the paths shown would be different.

Note that conda/miniconda gets setup with path names hardwired into its files, so you will not just be able to move your “miniconda3” directory and update these path locations later and still have it work.

Disable conda prompt string (opt)

I do want to have a conda environment up and running in each new terminal by default (so that Python is immediately available), but I don’t like having “(base)” appearing before my prompt all the time. To turn it off, you can just run:

conda config --set changeps1 False

and then in each new terminal, you won’t have that appear anymore. (To make your existing terminal recognize this change, source your shell’s ~/.*rc file, e.g. source ~/.bashrc or source ~/.cshrc.)

If in the future you want to re-enable this behavior, then you can always run:

conda config --set changeps1 True

Make basic conda environments

To see what conda environments are currently available on your computer, you can type:

conda env list

The name of all available environments will appear in the first column, as well as their locations in the second. The environment which is currently loaded will have an asterisk * after its name.

To create a new environment, at its simplest you can specify a Python version and a set of modules to install, such as:

conda create -y                 \
      -n py37_afni_tiny         \
      python=3.7                \
      matplotlib numpy

where the new environment’s name will be “py37_afni_tiny”; I called it this because that is basically the minimal set of modules used within AFNI (and even those aren’t used very often).

To make a similar setup for Python 2.7 (no earlier versions of Python should be used), one could run:

conda create -y                 \
      -n py27_afni_tiny         \
      python=2.7                \
      matplotlib numpy

Now, if I type conda list env, I will see a list of all my available environments:

# conda environments:
#
base                  *  /home/USERNAME/miniconda3
py27_afni_tiny           /home/USERNAME/miniconda3/envs/py27_afni_tiny
py37_afni_tiny           /home/USERNAME/miniconda3/envs/py37_afni_tiny

To turn off a Conda environment, you can run:

conda deactivate

To activate a particular environment, you can run conda activate NAME, such as:

conda activate py27_afni_tiny

To see what modules are installed in your active environment (and their version numbers) you can run:

conda list

… which, in the current “py27_afni_tiny” would be as follows (and you might have slightly different things):

- show list output y/n -

So, in this environment, I could run a program that imports matplotlib, whereas in the “base” environment, I couldn’t.

To add a module to an existing environment, you can use the following syntax:

conda install -n <env_name> <package>

So, for example example, you could add the scipy module to one of the above environments with:

conda install -n py27_afni_tiny scipy

Specify default environment for the terminal

From the Conda initialization, the “base” environment is the default one running in any new terminal. We might prefer to make one of our newly made environments the default. To do so, I will include a line conda activate NAME in my shell’s ~/.*rc file somewhere after the “conda initialize” lines.

Thus, since I am running “bash” shell, I have the following line in my ~/.bashrc file:

conda activate py37_afni_tiny

Once you source that ~/.*rc file again or open a new terminal, you should see that specific environment loaded (look for the asterisked env in the output of conda env list). If it didn’t work, please check that your conda version is at least 4.6 (via conda -V).

If you do choose to automatically activate your own env like this, then you might also want to run this in a terminal:

conda config --set auto_activate_base false

so that conda doesn’t pre-load the “base” environment unnecessarily (taking a bit of time).

Make conda environments, more generally

The environment builder works a bit like a package manager, where it can get a lot of common modules from a default, central repository, but if you want more specialized ones, you might have to add from another place. To add new repositories to pull from, you add a channel to your Conda setup.

Let’s say you want to add the Sphinx module with cloud-theme support (I doubt you will, but just as an example). If you try:

conda create -y                 \
    -n py37_afni_with_sph       \
    python=3.7                  \
    matplotlib numpy            \
    sphinx cloud_sptheme

You will likely get the following message:

- show text output y/n -
Collecting package metadata (current_repodata.json): done
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: failed

PackagesNotFoundError: The following packages are not available from current channels:

  - cloud_sptheme

Current channels:

  - https://repo.anaconda.com/pkgs/main/linux-64
  - https://repo.anaconda.com/pkgs/main/noarch
  - https://repo.anaconda.com/pkgs/r/linux-64
  - https://repo.anaconda.com/pkgs/r/noarch

To search for alternate channels that may provide the conda package you're
looking for, navigate to

    https://anaconda.org

and use the search bar at the top of the page.

This message: 1) tells us our current channels don’t contain this module; 2) shows us our current channels; and 3) helpfully directs us to a webpage to search for a new channel that might have it.

So, searching for “cloud_sptheme” at https://anaconda.org/, one of the top package owners appears to be “conda-forge” (and this is a fairly large platform). So, to add it to my repository list for getting modules, I would run:

conda config --add channels conda-forge

Then, I can retry my conda create .. command above, which should result in success this time.

Thus, if you try to build an environment and get told that some desired module can’t be found, you can search for it amongst available channels, add that channel to your Conda setup, and try again.

1.6.3. Set up Conda (quick)

  1. Download and install

    At ensuing prompts, I mostly type “yes” and/or accept default options; see verbose description above.

    • For Linux and tcsh:

      set script_file = Miniconda3-latest-Linux-x86_64.sh
curl -O https://repo.anaconda.com/miniconda/${script_file}
bash ${script_file}

      when done:

      source ~/.cshrc

    • For Linux and bash:

      script_file=Miniconda3-latest-Linux-x86_64.sh
curl -O https://repo.anaconda.com/miniconda/${script_file}
bash ${script_file}

      when done:

      source ~/.bashrc

    • For Mac and tcsh:

      set script_file = Miniconda3-latest-MacOSX-x86_64.sh
curl -O https://repo.anaconda.com/miniconda/${script_file}
bash ${script_file}

      when done:

      source ~/.cshrc

    • For Mac and bash:

      script_file=Miniconda3-latest-MacOSX-x86_64.sh
curl -O https://repo.anaconda.com/miniconda/${script_file}
bash ${script_file}

      when done:

      source ~/.bashrc

  2. Remove annoying prompt string (opt)

    conda config --set changeps1 False

  3. Add any channels, if needed (opt)

    Many standard modules are default channels, but one can add more as necessary, e.g.:

    conda config --add channels conda-forge --add channels anaconda

    To find out what channel has your module of interest, use the searchbar at the top of this page: https://anaconda.org

  4. Make some new environments

    conda create -y                 \
      -n py37_afni_tiny         \
      python=3.7                \
      matplotlib numpy

conda create -y                 \
      -n py27_afni_tiny         \
      python=2.7                \
      matplotlib numpy

  5. Add to an existing environment

    conda install -n py27_afni_tiny scipy

  6. Activate an env by default

    Specify the env to activate in your ~/.*rc file.

    Open up the ~/.bashrc or ~/.cshrc text file and put conda activate NAME after the conda-initialization lines in that file, or, e.g. copy+paste:

    • For tcsh:

      echo "" >> ~/.cshrc
echo "conda activate py37_afni_tiny" >> ~/.cshrc
echo "" >> ~/.cshrc

    • For bash:

      echo "" >> ~/.bashrc
echo "conda activate py37_afni_tiny" >> ~/.bashrc
echo "" >> ~/.bashrc

    Note: This assumes your conda version (conda -V) is at least 4.6.

    If you do automatically activate your own env, then you might also want to do this:

    conda config --set auto_activate_base false

    so that conda doesn’t pre-load the “base” environment unnecessarily (taking a bit of time).

  7. Quicktasks with Conda

    List modules (starred/asterisked one is active):

    conda env list

    Deactivate current module:

    conda deactivate

    Activate/switch to a specific module:

    conda activate NAME

    See module+version list in current env:

    conda list

1.6.4. Fancier things with Conda

There are a lot of fancy things that can be done with Conda that we will not describe here. A good starting point is the Managing Environments documentation.

Cloning

One concept with Conda is cloning environments: if I can setup a Conda environment on my laptop with a certain set of modules, each with a certain version number, then I can “clone” it and use that exact recipe to setup a duplicate environment on a different computer. This is a nice concept for reproducibility (as sometimes using different version numbers of modules can affect outputs/results).

More on cloning and building identical conda envs can be read.

Note that in practice, truly duplicating environments exactly is actually pretty tough. Getting very close might be good enough for most purposes, though, in practice.

1.6.5. A note on making envs and choosing modules

It is entirely up to you, Dear User, what modules you install and how you organize your environments (and if you even choose to use Conda). At the moment AFNI-land has pretty minimal Python requirements. In fact, the AFNI set of recommended modules might simply fit inside those requirements that you have for other software/uses, and you might not need to do anything new.

We certainly don’t anticipate or desire a person to set up one specific environment for running AFNI, then another for running some other software, and then another for another project… While that is possible, it seems annoying and inefficient. And unnecessary. So, hopefully, you can set up one environments (or maybe two, because of the Python-2.7 and Python-3.* split), and not have to switch too much.