Skip to contents

Python alongside R

Source: vignettes/articles/python.qmd

Experimental: this may still change, see #108.

A Pixi environment can have Python and its packages next to R and its packages, in the same pixi.lock, so a collaborator gets the same Python stack too. R uses it through reticulate.

Set it up

or, for a new project, use_pixi(python = TRUE). This adds Python, reticulate, and what reticulate needs to use the environment’s Python, and points reticulate at it. Restart R afterwards.

Then add Python packages. Most are on conda-forge:

pixi_add("conda::numpy")

For those that aren’t, such as pydytuesday (TidyTuesday’s data, for Python), add them from PyPI:

pixi_add("pypi::pydytuesday")

Use them from R as usual:

numpy <- reticulate::import("numpy")

What it sets up

  • RETICULATE_PYTHON, in pixi.toml’s [activation.env], points reticulate at the environment’s Python. Without it, reticulate makes a Python of its own, which doesn’t have the project’s packages. It applies to R started with pixi run, and to R started by an IDE, through the project’s .Rprofile.
  • micromamba and RETICULATE_CONDA: reticulate activates a conda environment again before using its Python, with a conda program, which a Pixi project doesn’t have (rstudio/reticulate#1650). conda-forge’s micromamba does that.
  • libpython, for Python 3.15 or later: conda-forge keeps Python’s shared library, which reticulate loads, in a package of its own.
  • On Windows, the settings are written differently, as Pixi doesn’t expand $CONDA_PREFIX there. They’re only written if the project has win-64; if you add it later, run use_pixi_python() again.

Things to know

  • reticulate::py_require() doesn’t install into the project’s Python: add Python packages with pixi_add() instead.
  • PyPI packages are for what conda-forge doesn’t have: a package from conda-forge can bring the system libraries it needs, one from PyPI can’t. They need Python in the environment, and Pixi installs them with uv, next to the conda packages.