Reticulate

A simple reference on using the reticulate package within R. Things make sense in real time and then coming back to it later and I’ve lost all memory of what I’ve done or how I’ve done it.

Assuming that miniconda has been installed.

Creating a Conda environment

Create new Rstudio project and then within that project create a new conda environment

library(reticulate)
conda_create("twg-site")

Reticulate seems to have an aggressive habit of forcing a specific python version within the environment. And what’s interesting, is that I don’t have an environmental variable that points to a specific python version. But if I use the Sys.getenv('RETICULATE_PYTHON') function for it I find that the python version is set somewhere:

So, when we call py_config() to see which version is being used, it’s always this base python3, which isn’t that helpful.

library(reticulate)
reticulate::py_config()

## python:         /opt/homebrew/bin/python3
## libpython:      /opt/homebrew/opt/python@3.11/Frameworks/Python.framework/Versions/3.11/lib/python3.11/config-3.11-darwin/libpython3.11.dylib
## pythonhome:     /opt/homebrew/Cellar/python@3.11/3.11.3/Frameworks/Python.framework/Versions/3.11:/opt/homebrew/Cellar/python@3.11/3.11.3/Frameworks/Python.framework/Versions/3.11
## version:        3.11.3 (main, Apr  7 2023, 20:13:31) [Clang 14.0.0 (clang-1400.0.29.202)]
## numpy:           [NOT FOUND]
## 
## NOTE: Python version was forced by RETICULATE_PYTHON_FALLBACK

If we use the conda_list() function we can see the various conda environments that have been set up. So what we want to do is create our own .Renviron file within our project to ensure that the proper environment is called every time. This way, we know that packages downloaded for a specific project can be found again.

conda_list()

##           name                                                                    python
## 1         base                   /Users/taylorgrant/Library/r-miniconda-arm64/bin/python
## 2 r-reticulate /Users/taylorgrant/Library/r-miniconda-arm64/envs/r-reticulate/bin/python
## 3     scrapper     /Users/taylorgrant/Library/r-miniconda-arm64/envs/scrapper/bin/python
## 4   topicmodel   /Users/taylorgrant/Library/r-miniconda-arm64/envs/topicmodel/bin/python

Now create an .Renviron file with the proper directory location

usethis::edit_r_environ(scope = 'project')

# this opens the new file then add this, save, and restart
RETICULATE_PYTHON=/Users/taylor_grant/Library/r-miniconda/envs/twg-site/bin/python

On restart, the correct conda environment will be found.

Installing packages

Installing packages is very easy via the conda_install() function. There are essentially three arguments - the conda environment, e.g., “twg-site”, the package to install, and whether to use pip or the conda channels. By default, it uses conda channels.

reticulate::conda_install("twg-site", "statsmodels", pip = TRUE)

You can specify specific package versions to install by including the package version, e.g., conda_install("twg-site", top2vec=="1.0.26", pip = TRUE)

Removing packages

Just as easy, use the conda_remove() function.

reticulate::conda_remove("twg-site", "statsmodels", pip = TRUE)

Occasionally, the conda_remove() function will fail to find a package that is clearly installed in the conda environment. If that happens, the following code is useful. Put the package name in quotes. The trailing ‘-y’ tells not to ask for confirmation.

system2(reticulate::py_exe(), c("-m", "pip", "uninstall", <package>, '-y))

Working in Python

To switch to python, simply call

reticulate::repl_python()

You can exit the environment by typing exit in the command line.

Within R you can call objects created in python with by specifying py$ much like you would call a column from a data frame.

For the rest see calling_python