RStudio + R + Python Reproducible Environments

Setting Up Environments

Locally

Check that you have installed R and Anaconda. For R make sure you have a version you want selected in RStudio (Tools > Global Options > R version)

Create new RStudio project: RStudio -> new project -> new directory

On the Cluster

Make anaconda available in the project. The path you specify depends on which anaconda version in the system you are using.
In order to do that, add the following line into .Rprofile file. Close and open R project.

Now you can specify you are going to use python packages, and that those packages will be handled with anaconda. The following command will create conda environment withing renv directory

Installing Packages and Making Snapshots

• Install R packages as usual

• Install python packages using.  Remember to change cache directory if applicable (as described in renv page).
  
  

Make snapshot of current package versions

snapshot command will create two files in the project directory

• renv.lock - This file will contain info about all the R packages used, R version info, info on conda env

• environment.yml - this file will contain info about all the conda packages installed.

Commit those files to git together with your script files

Restoring an Environment

Assuming you have renv.lock and environment.yml file in the project directory you would use `renv::restore()` command. This will install R packages and conda packages (including python itself). You need to make sure that R version you use is the same as the one in renv.lock

This should work nicely as long as you restore at the same OS, at which renv::snapshot() was created (Windows, Linux, macOS)

Restoring on a Different OS

In case you or your collaborators are going to reproduce the same environment on a different OS (for example Widnows -> Linux or Linux -> Windows), you may need to change content of this file. In particular

1. Recreate environment.yml file using commands: 

This will overwrite the one created by renv::snapshot(), which does include build information, which is specific to OS, and may create problems for environment setup on another OS

2. change name parameter within the file to null

3. change prefix parameter within file to appropriate value

• For Linux and MacOS
  ./renv/python/condaenvs/renv-python

• For Windows
  .\renv\python\condaenvs\renv-python

4. Some conda packages are only available on Linux or only on Windows.
   While executing restoration command renv::restore() you may see that some conda packages are not available. One of the most likely reasons is that those packages are specific to operating system. Examples: vs2015, ucrt, vc are Windows specific; libnsl, libgfortran5 are Linux specific. Thus, you may need to delete those lines from the file for renv::restore() to work properly

Using with RStudio on Cluster

You can use Open OnDemand, open RStudio and create RStudio project the same way you would do locally: either by creating a project in a new directory, or by cloning from the GitHub repository. 

Additional Reading

These pages may be useful:

• https://mpn.metworx.com/packages/renv/0.12.5/reference/use_python.html

• https://rstudio.github.io/reticulate/articles/python_packages.html