Python and R with Anaconda

Python and R with Anaconda#

To support diverse Python and R workflows, Research Computing users can utilize Anaconda. Anaconda provides the Conda package manager, which allows for easy installation of software and associated packages. The Conda package manager provides support for Python, R, and many other application stacks. CURC also supports Mamba, an alternative package manager that allows parallel downloading of repository data and package files using multi-threading.

The following documentation describes how to activate the CURC Anaconda distribution and our default environments, as well as how to create and activate your own custom Anaconda environments. For more information on utilizing Mamba (an alternative package manager) please see the section Mamba Package Manager below. Additional documentation on CURC OpenOnDemand is available for users desiring to interact with their custom environments via Jupyter notebooks.

Configuring Conda and Mamba with .condarc#

The Conda and Mamba package managers allow modification of default settings. These settings are specified in a text file called .condarc. For possible configuration options, please see Using the .condarc conda configuration file. The .condarc file should exist within a user’s /home/$USER directory and can be quickly accessed using the file’s full path at ~/.condarc. A .condarc file is important because by default Conda and Mamba will put all package source code and environments in your /home/$USER directory. This quickly becomes an issue because your /home/$USER directory has limited storage capacity (see The Home Filesystem section). For this reason, it is highly suggested that user’s redirect material to /projects/$USER.

Important

When loading the Anaconda or Mamba modules, a .condarc file will be created for you in your /home/$USER directory, if the file does not exist. If the file is created for you, it will contain the following content, which places Conda and Mamba items in your /projects/$USER directory:

pkgs_dirs:
  - /projects/$USER/.conda_pkgs
envs_dirs:
  - /projects/$USER/software/anaconda/envs

Although we will automatically create this file for you when you load the associated modules, you may want to modify .condarc. This can be done by opening your .condarc file in your favorite text editor (e.g., nano, vim) and modifying it.

[johndoe@sc3cpu-a7-u19-1 ~]$ nano ~/.condarc

After making changes, save and exit the file. Any modifications made and saved will be permanent unless you modify .condarc later.

Important

CSU and XSEDE/ACCESS users may need to use a custom $USER variable because the @ symbol in the usernames can occasionally be misinterpreted by environments that employ PERL. Directions to set up a custom user variable can be found under our CSU and XSEDE username documentation.

Using the CURC Anaconda environment#

Follow these steps from a Research Computing terminal session on an Alpine acompile node or within a Alpine/Blanca batch or interactive job.

Activate the CURC Anaconda environment:#

Run the following command to load the base Anaconda software:

[johndoe@c3cpu-a7-u19-1 ~]$ module load anaconda

Note

The command above activates the base environment for python3, which as of 2020 is the only supported python standard. For users requiring legacy python2, you can still use Conda to create a custom environment with the python2.X version of your choice (we provide an example of how to do this below)

You will know that you have properly activated the environment because you should see (base) in front of your prompt. For example:

(base) [johndoe@c3cpu-a7-u19-1 ~]$

Using Conda:#

Now that you have activated the base Conda environment, you can use Conda for Python and R! There are two ways forward, depending on your needs. You can:

1. Use one of CURC’s pre-installed environments.

  • Pros: You can begin using one of these immediately, and they contain mainy of the most widely used python and R packages.

  • Cons: These are root-owned environments, so you can not add additional packages.

or

2. Create your own custom environment(s).

  • Pros: You own these, so you can add packages as needed, control package versions, etc.

  • Cons: There really aren’t any cons, other than the time needed to create a custom environment (usually 5-30 minutes depending on the number of packages you install).

Both options are discussed below.

Using one of CURC’s pre-installed environment:#

To use the CURC environment with OpenMP, run the following command with Anaconda initialized:

(base) [johndoe@c3cpu-a7-u19-1 ~]$ conda activate 
/curc/sw/anaconda3/2022.10/envs/pyomp_2022

You will know that you have properly activated the environment because you should see (pyomp_2022) replace the (base) in front of your prompt.

To see the python packages available in the environment, you can type conda list.

Similarly, to use the CURC R distribution (R v3.6.0), run the following command with Anaconda initialized:

(base) [johndoe@c3cpu-a7-u19-1 ~]$ conda activate 
/curc/sw/anaconda3/2019.03/envs/rstudio

You will know that you have properly activated the environment because you should see (rstudio) in front of your prompt. To see the R packages available in the environment, you can type conda list. Now, you can use R as you normally would.

Because interactive development is more easily done locally, most CURC R users exclusively run R code within batch jobs. Should you need to use rstudio on top of R for interactive development on Alpine, you can login to our system with X11-forwarding (ssh -X) and initiate an rstudio session from within an interactive job.

Create your own custom environment:#

Tip

In the examples below the environment is created in /projects/$USER/software/anaconda/envs, which is specified under envs_dirs in your .condarc file. Environments can be installed in any user-writable location the user chooses; just add the path to ~/.condarc.

1. Initialize Anaconda if you haven’t already done so:

[johndoe@c3cpu-a7-u19-1 ~]$ module load anaconda
(base) [johndoe@c3cpu-a7-u19-1 ~]$

2. Create a custom environment:

Here we create a new environment called mycustomenv (you can call it anything you want!)

(base) [johndoe@c3cpu-a7-u19-1 ~]$ conda create -n mycustomenv

If you want a specific version of python or R, you can modify the above command as follows (e.g.):

Python v2.7.16:

(base) [johndoe@c3cpu-a7-u19-1 ~]$ conda create -n mycustomenv python==2.7.16

Python v3.6.8:

(base) [johndoe@c3cpu-a7-u19-1 ~]$ conda create -n mycustomenv python==3.6.8

Latest version of R:

(base) [johndoe@c3cpu-a7-u19-1 ~]$ conda create -n mycustomenv r-base

3. Activate your new environment:

(base) [johndoe@c3cpu-a7-u19-1 ~]$ conda activate mycustomenv
(mycustomenv) [johndoe@c3cpu-a7-u19-1 ~]$

If successful, your prompt will now be preceded with (mycustomenv).

4. Install needed packages in your new environment:

The best way to do this for python packages is to install everything you need with one command, because it forces conda to resolve package conflicts. For example:

(mycustomenv) [johndoe@c3cpu-a7-u19-1 ~]$ conda install numpy scipy tensorflow

For R packages, it is easiest to start an R session and then install the packages as you normally would with “install.packages”. For example:

(mycustomenv) [johndoe@c3cpu-a7-u19-1 ~]$ R
>install.packages("ggplot2")

If you encounter a --- Please select a CRAN mirror for use in this session --- message, you can select a US mirror from the provided list or use the repos install flag:

>install.packages('RMySQL', repos='http://cran.us.r-project.org')

For more information on managing conda environments, check out Anaconda’s documentation here.

Basic Conda commands to get you started:#

Command

Function

conda list

List the packages currently installed in the environment

conda search <package>

Searches the Anaconda package channel for a package named <pakage>

conda install <package>

Installs a package named <package> to your currently loaded environment

conda uninstall <package>

Uninstalls a package named <package> from your currently loaded environment

conda env list

List the Conda environments currently available

conda create <env>

Creates a new anaconda environment named <env>

conda remove --name <env> --all

Removes an environment named <env>

conda deactivate

Deactivates current environment

Troubleshooting#

If you are having trouble loading a package, you can use conda list or pip freeze to list the available packages and their version numbers in your current Conda environment. Use conda install <package> to add a new package or conda install <package==version> for a specific version; e.g., conda install numpy=1.16.2.

Sometimes Conda environments can “break” if two packages in the environment require different versions of the same shared library. In these cases you try a couple of things.

  • Reinstall the packages all within the same install command (e.g., conda install <package1> <package2>). This forces Conda to attempt to resolve shared library conflicts.

  • Create a new environment and reinstall the packages you need (preferably installing all with the same conda install command, rather than one-at-a-time, in order to resolve the conflicts).

Mamba Package Manager#

Mamba is a fast, robust, and cross-platform package manager that aims to be a drop-in replacement for Conda. Utilizing Mamba can improve the speed and reliability of constructing an environment. To use Mamba on an Alpine or Blanca compute node, perform the following module load:

[johndoe@c3cpu-a7-u19-1 ~]$ module load mambaforge

The command above activates the base environment provided by Mamba. You will know that Mamba has been correctly loaded once you see (base) in front of your prompt. For example:

(base) [johndoe@c3cpu-a7-u19-1 ~]$

Once Mamba has been properly loaded, you can utilize almost all core command and configuration options available to Conda. For commands, this can be done by replacing conda with mamba. For example:

[johndoe@c3cpu-a7-u19-1 ~]$ mamba create -n mycustomenv

Note

Mamba will utilize configurations specified in .condarc. For more information on the .condarc file, see Configuring Conda and Mamba with .condarc above.

Dbus Error#

If you see a ‘dbus’ connection error when trying to connect via a virtual environment:

Could not connect to session bus: Failed to connect to socket /tmp/dbus-oBg2HbRfLi: Connection refused.

This is likely due to your ~/.bashrc configuration file auto-activating a Conda environment with a problematic dbus package. You can resolve this issue by opening your ~/.bashrc with a text editor (ex. vim, nano) and commenting out the following lines (or any lines that add a Conda environment to your $PATH):

Tip

Commenting lines out instead of removing them will allow you to add them back in later if needed. These lines have been commented out using # preceding each line.

# >>> conda initialize >>>
# !! Contents within this block are managed by 'conda init' !!
# __conda_setup="$('/curc/sw/anaconda3/2019.07/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
# if [ $? -eq 0 ]; then
#     eval "$__conda_setup"
# else
#     if [ -f "/curc/sw/anaconda3/2019.07/etc/profile.d/conda.sh" ]; then
#         . "/curc/sw/anaconda3/2019.07/etc/profile.d/conda.sh"
#     else
#         export PATH="/curc/sw/anaconda3/2019.07/bin:$PATH"
#     fi
# fi
# unset __conda_setup
# <<< conda initialize <<<

Keep in mind that doing this means Conda is not automatically sourced by your ~/.bashrc so you will have to manually source the base Conda environment with module load anaconda to activate the base environment.

Contents