brmspy 0.1.12

Python-first access to R's brms with proper parameter names, ArviZ support, and cmdstanr performance

brmspy

Python-first access to R's brms with proper parameter names, ArviZ support, and cmdstanr performance. The easiest way to run brms models from Python.

This is an early development version of the library, use with caution.

Github repo and issues

Installation

pip install brmspy

First-time setup (installs brms, cmdstanr, and CmdStan in R):

from brmspy import brms
brms.install_brms() # requires R to be installed already

Prebuilt Runtimes (Optional)

For faster installation (~20-60 seconds vs 20-30 minutes), use prebuilt runtime bundles:

from brmspy import brms
brms.install_brms(use_prebuilt_binaries=True)

Windows RTools

In case you don't have RTools installed, you can use the flag install_rtools = True. This is disabled by default, because the flag runs the full rtools installer and modifies system path. Use with caution!

from brmspy import brms
brms.install_brms(
    use_prebuilt_binaries=True,
    install_rtools=True # works for both prebuilt and compiled binaries.
)

System Requirements

R >= 4.0

Linux (x86_64):

• glibc >= 2.27 (Ubuntu 18.04+, Debian 10+, RHEL 8+)
• g++ >= 9.0

macOS (Intel & Apple Silicon):

• Xcode Command Line Tools: xcode-select --install
• clang >= 11.0

Windows (x86_64):

• Rtools 4.0+ with MinGW toolchain
• g++ >= 9.0

Download Rtools from: https://cran.r-project.org/bin/windows/Rtools/

Quick Start

from brmspy import brms, prior
import arviz as az

# Load data
epilepsy = brms.get_brms_data("epilepsy")

# Fit model
model = brms.fit(
    formula="count ~ zAge + zBase * Trt + (1|patient)",
    data=epilepsy,
    family="poisson",
    priors=[
        prior("normal(0, 1)", "b"),
        prior("exponential(1)", "sd", group="patient"),
        prior("student_t(3, 0, 2.5)", "Intercept")
    ],
    chains=4,
    iter=2000
)

# Analyze
az.summary(model.idata)
az.plot_posterior(model.idata)

Key Features

• Proper parameter names: Returns b_Intercept, b_zAge, sd_patient__Intercept instead of generic names like b_dim_0
• arviz integration: Returns arviz.InferenceData by default for Python workflow
• brms formula syntax: Full support for brms formula interface including random effects
• Dual access: Results include both .idata (arviz) and .r (brmsfit) attributes
• No reimplementation: Delegates all modeling logic to real brms. No Python-side reimplementation, no divergence from native behavior. Opinionated wrappers that rebuild formulas or stancode in Python inevitably drift from brms and accumulate their own bugs.
• Prebuilt Binaries: Fast installation with precompiled runtimes containing cmdstanr and brms (50x faster, 25 seconds on Google Colab)

API Reference

brmspy documentation brms documentation

Setup Functions

• brms.install_brms() - Install brms, cmdstanr, and CmdStan
• brms.get_brms_version() - Get installed brms version

Data Functions

• brms.get_brms_data() - Load example datasets from brms
• brms.save_rds() - Save brmsfit or another robject
• brms.load_rds_fit() - Load saved brmsfit object as FitResult (with idata)
• brms.load_rds_raw() - Load r object

Model Functions

• brms.formula() - Define formula with kwargs
• brms.fit() or brms.brm() - Fit Bayesian regression model
• brms.summary() - Generate summary statistics as DataFrame
• brms.make_stancode() - Generate Stan code for model

Prior functions

• brms.prior() - Define a prior with same syntax as r-s prior_string
• brms.get_prior() - Get pd.DataFrame describing default priors
• brms.default_prior() - Get pd.DataFrame describing default priors

Families functions

• brms.family() - Get family object of FitResult
• brms.brmsfamily() - Construct family object from kwargs
• brms.families.gaussian(), ...bernoulli(), ...beta_binomial(), etc - Wrappers around brmsfamily for faster family object construction

Prediction Functions

• brms.posterior_epred() - Expected value predictions (without noise)
• brms.posterior_predict() - Posterior predictive samples (with noise)
• brms.posterior_linpred() - Linear predictor values
• brms.log_lik() - Log-likelihood values

Usage

Basic Model

from brmspy import brms

kidney = brms.get_brms_data("kidney")

model = brms.fit(
    formula="time ~ age + disease",
    data=kidney,
    family="gaussian",
    chains=4,
    iter=2000
)

With Priors

from brmspy import prior

model = brms.fit(
    formula="count ~ zAge + (1|patient)",
    data=epilepsy,
    family="poisson",
    priors=[
        prior("normal(0, 0.5)", "b"),
        prior("cauchy(0, 1)", "sd")
    ],
    chains=4
)

Model Summary

from brmspy import summary

# Get summary statistics as DataFrame
summary_df = summary(model)
print(summary_df)

Predictions

# Expected value (without noise)
epred = brms.posterior_epred(model, newdata=new_data)

# Posterior predictive (with noise)
ypred = brms.posterior_predict(model, newdata=new_data)

# Linear predictor
linpred = brms.posterior_linpred(model, newdata=new_data)

# Log likelihood
loglik = brms.log_lik(model, newdata=new_data)

Access Both Python and R Objects

model = brms.fit(formula="y ~ x", data=data, chains=4)

# Python workflow with arviz
az.summary(model.idata)
az.plot_trace(model.idata)

# R workflow (if needed)
import rpy2.robjects as ro
ro.r('summary')(model.r)

Sampling Parameters

model = brms.fit(
    formula="y ~ x + (1|group)",
    data=data,
    iter=2000,      # Total iterations per chain
    warmup=1000,    # Warmup iterations
    chains=4,       # Number of chains
    cores=4,        # Parallel cores
    thin=1,         # Thinning
    seed=123        # Random seed
)

Requirements

Python: 3.10-3.14

R packages (auto-installed via brms.install_brms()):

• brms >= 2.20.0
• cmdstanr
• posterior

Python dependencies:

• rpy2 >= 3.5.0
• pandas >= 1.3.0
• numpy >= 1.20.0
• arviz (optional, for InferenceData)

Development

git clone https://github.com/kaitumisuuringute-keskus/brmspy.git
cd brmspy
./init-venv.sh
pytest tests/ -v

Architecture

brmspy uses:

• brms::brm() with cmdstanr backend for fitting (ensures proper parameter naming)
• posterior R package for conversion to draws format
• arviz for Python-native analysis and visualization
• rpy2 for Python-R communication

Previous versions used CmdStanPy directly, which resulted in generic parameter names. Current version calls brms directly to preserve brms' parameter renaming logic.

License

Apache License 2.0

Credits

• Original concept: Adam Haber
• Current maintainer: Remi Sebastian Kits
• Built on brms by Paul-Christian Bürkner