A reactive notebook for Python — run reproducible experiments, query with SQL, execute as a script, deploy as an app, and version with git. Stored as pure Python. All in a modern, AI-native editor.
Python artificial-intelligencedagdata-sciencedata-visualizationdataflowdeveloper-toolsmachine-learningnotebookspipelinepythonreactivesqlweb-app
# /// script
# requires-python = ">=3.12,<3.13"
# dependencies = [
# "marimo",
# "polars>=1",
# "pyarrow>=17",
# "rpy2>=3.6",
# "rpy2-arrow==0.1.3",
# ]
#
# [tool.pixi.workspace]
# channels = ["conda-forge"]
#
# [tool.pixi.dependencies]
# r-base = "*"
# r-dplyr = "*"
# r-arrow = ">=17"
# ///
import marimo
__generated_with = "0.25.2"
app = marimo.App()
with app.setup:
import os
# Load R from the Pixi environment at runtime.
os.environ["RPY2_CFFI_MODE"] = "ABI"
import marimo as mo
import polars as pl
import rpy2.robjects
import rpy2_arrow.arrow
@app.cell(hide_code=True)
def _():
mo.md("""
# R with Pixi
With Pixi sandboxing, a marimo notebook can describe more than its Python
dependencies, including system libraries and even another language's runtime.
This demo declares R and Python dependencies together, uses marimo's
reactivity to run R code when a slider changes, and shares the resulting
Arrow table back to Python without copying its data.
""")
@app.cell(hide_code=True)
def _():
mo.md("""
With [Pixi installed](https://pixi.prefix.dev/latest/installation/), run:
```bash
marimo edit --sandbox=pixi examples/misc/pixi_r.py
```
""")
@app.cell
def _():
min_len = mo.ui.slider(
start=4.0,
stop=8.0,
step=0.1,
value=5.0,
label="Minimum sepal length (cm)",
)
min_len
return (min_len,)
@app.cell
def _(min_len):
r_table = rpy2.robjects.r(f"""
iris |>
dplyr::filter(Sepal.Length >= {min_len.value}) |>
dplyr::mutate(Species = as.character(Species)) |>
arrow::arrow_table()
""")
df = pl.from_arrow(rpy2_arrow.arrow.rarrow_to_py_table(r_table))
df
if __name__ == "__main__":
app.run()