rpynative

Use R packages natively in Python, no R knowledge required.

rpynative lets Python developers load and use R packages as if they were normal Python packages. No R syntax, no manual object conversion, and unlike other R-Python bridges, rpynative even supports real IDE autocomplete for R functions.

Why not just use rpy2?

rpynative is built on top of rpy2, rpy2 is the engine, rpynative is what a Python-only developer actually experiences using it. rpy2 exposes R concepts directly (R vectors, manual converter contexts, raw R objects), it's a bridge for people who already know R. rpynative is built for people who don't, and want R's tools to feel like native Python.

Specifically, rpynative adds:

• Zero R knowledge required : pass plain Python lists/DataFrames, get back plain Python dicts/DataFrames
• Auto-install : missing R packages are installed from CRAN automatically
• Preserved native reporting : R's own nicely-formatted print output (e.g. a package's custom statistical report) is captured and shown, not just raw data
• Object-chaining and generic dispatch handled transparently: things like ggplot2's + syntax, or predict() working across any model type, just work
• Auto-generated IDE stubs: real autocomplete and parameter hints in editors like VS Code for a dynamically loaded R package's functions
• Round-tripping of complex objects: trained models, plots, and other R objects can be passed back into other R functions without losing their identity

Example

import rpynative

stats = rpynative.load("stats")
result = stats.mean([2, 4, 6, 8, 10])
print(result)  # 6.0

Real statistical reports, not raw R objects

statease = rpynative.load("statease")
result = statease.ttest_interpret([88, 92, 79, 85, 90], [70, 65, 72, 68, 74])

print(result)
# -- statease T-Test Report ----------------------------------------
#   Test         : Independent Samples T-Test
#   ...

print(result['p_val'])  # still works like a normal dict

Formulas, DataFrames, matrices, and S4 objects

import pandas as pd

data = pd.DataFrame({"score": [...], "group": [...]})
result = statease.anova_interpret("score ~ group", data=data)
print(result)

Visual/plotting packages

ggplot2 = rpynative.load("ggplot2")

plot = ggplot2.ggplot(data, ggplot2.aes(x="x", y="y")) + ggplot2.geom_point()
ggplot2.ggsave("plot.png", plot=plot)

Real machine learning workflows

randomForest = rpynative.load("randomForest")

model = randomForest.randomForest("outcome ~ feature1 + feature2", data=data)
predictions = randomForest.predict(model, new_data)

IDE autocomplete for R functions

from rpynative.core import load_typed
from rpynative.stubs.statease import StateaseStub

statease = load_typed("statease", StateaseStub)
statease.  # <- VS Code shows real autocomplete here

Note: autocomplete currently requires generating a stub once per package (rpynative.stub_generator.generate_stub("packagename")). Automatic stub generation on load() is a planned improvement.

Features

• Load any R package with one line, auto-installs it from CRAN if missing
• Automatic conversion: lists, numbers, pandas DataFrames, R formulas, matrices (as numpy arrays), S4 objects (as dicts)
• Rich report output preserved alongside clean programmatic access
• Object-chaining support (+ operator) for packages like ggplot2
• Generic function dispatch (predict, summary, etc.) works across any model type
• Built-in R datasets accessible directly as pandas DataFrames
• Auto-generated IDE stubs for real autocomplete
• Clean Python errors instead of raw R tracebacks

Requirements

• Python 3.9+
• R installed and available on your system (R packages are auto-installed as needed)

Installation

pip install rpynative

Known limitations

• IDE autocomplete requires a manual one-time stub generation step per package
• Not yet tested against every R object type across all ~20,000 CRAN packages, common types (S3, S4, matrices, data.frames, formulas) are covered and tested; more exotic types may surface new edge cases
• Windows, macOS, and Linux should all work (R + rpy2 support all three), but development and testing so far has been primarily on Windows

Status

Actively developed. Core engine, report handling, plotting support, matrix/S4 support, and IDE autocomplete are all working and tested against real published R packages, including real machine learning workflows.

Author

Built by DevWebWacky, also the author of the statease R package used throughout development and testing, and triageR

License

MIT