ArviZ

1.0.0 · active · verified Thu Apr 09

ArviZ is a Python package for exploratory analysis of Bayesian models. It provides tools for diagnostics, visualization, and inference data management, working with various probabilistic programming frameworks. Version 1.0.0 is the current major release, and the project has a regular, active release cadence.

Warnings

breaking ArviZ v1.0.0 and newer require Python 3.12 or higher. Users on older Python versions (e.g., 3.10, 3.11) must upgrade their Python environment to use ArviZ 1.x.
Fix: Upgrade your Python environment to 3.12+ or use an older ArviZ version compatible with your Python (e.g., `pip install arviz<1.0`).
breaking The `rpy2` dependency and the `az.from_rpy2` function were removed in ArviZ v1.0.0 due to the internal refactor into a meta-package structure.
Fix: Migrate any workflows relying on direct R-to-Python conversion via `from_rpy2` to alternative data transfer methods, or use an older ArviZ version if this functionality is critical.
gotcha Default values for several key plotting and analysis functions changed in ArviZ v0.13.0. For example, `az.plot_forest`'s `combined` argument, `az.plot_posterior`'s `kind` argument, and `az.loo`'s `pointwise` argument.
Fix: Review release notes for v0.13.0 and explicitly specify arguments like `combined`, `kind`, `pointwise` if you relied on previous default behaviors to ensure consistent output.

Install

pip install arviz Install ArviZ

Imports

arviz
```
import arviz as az
```
Commonly imported with the alias `az` for brevity and convention.
InferenceData
```
import arviz as az
idata = az.InferenceData(...)
```
The central data structure for ArviZ, usually accessed via `az` alias.

Quickstart

This quickstart demonstrates how to create a basic `InferenceData` object (often loaded from a model output) and use ArviZ to plot a trace and generate summary statistics. It uses `numpy` for data simulation and `arviz` for analysis and visualization.

import arviz as az
import numpy as np

# Simulate some data for demonstration
np.random.seed(42)
data = {
    'theta': np.random.normal(0, 1, size=1000),
    'phi': np.random.normal(5, 2, size=1000)
}

# Create a dummy InferenceData object (usually loaded from a model)
id = az.InferenceData(
    posterior=az.dict_to_dataset(data),
    sample_stats=az.dict_to_dataset({'lp': np.random.normal(0, 1, size=1000)})
)

# Plot a trace of the posterior samples
az.plot_trace(id, var_names=['theta'])

# Print summary statistics
print(az.summary(id, var_names=['theta']))

view raw JSON →