Installation

PyVBMC is available via pip and conda-forge and requires Python 3.10 or newer.

Basic installation

Install or upgrade PyVBMC with pip:

python -m pip install --upgrade pyvbmc

Or install it with Conda:

conda install --channel=conda-forge pyvbmc

PyVBMC 1.5 requires NumPy 2.0, SciPy 1.15 and matplotlib 3.9 or newer, and its conda-forge package requires Python 3.11 or newer. In an environment that holds an older version of one of them, conda can install PyVBMC 1.0.4 instead, without a warning: ask it for "pyvbmc>=1.5", or see the FAQ.

To learn whether a newer release exists and how to update, see the FAQ.

Optional integrations

Torch

The to_torch() export and the SVBMC class (stacking the posteriors of several VBMC runs) require torch 2.7 or newer. For a CPU-only installation, install torch from its official CPU wheel index first, then install the PyVBMC extra:

python -m pip install torch --index-url https://download.pytorch.org/whl/cpu
python -m pip install --upgrade "pyvbmc[torch]"

Installing torch first ensures that pip selects a CPU wheel. If your model or export will run on an accelerator, follow the torch installation selector for your platform before installing pyvbmc[torch].

With Conda, install the named packages directly:

conda install --channel=conda-forge "pyvbmc>=1.5" pytorch

ArviZ

The to_arviz() export uses the current ArviZ DataTree format and requires Python 3.12 or newer. Install it with:

python -m pip install --upgrade "pyvbmc[arviz]"

PyVBMC itself continues to support Python 3.10 and newer. On Python 3.10 or 3.11, pip does not install the ArviZ dependencies because of their Python version marker, and to_arviz is unavailable. With Conda, install the corresponding packages directly:

conda install --channel=conda-forge "pyvbmc>=1.5" arviz arviz-base

PyMC

The PyMCTarget adapter requires Python 3.12 or newer. Install PyMC and the ArviZ data interface with:

python -m pip install --upgrade "pyvbmc[pymc]"

The adapter is tested with PyMC 6.3.2, PyTensor 3.3.1–3.3.3 and ArviZ 1.3.0, using PyTensor’s default Numba linker on Windows and Linux. PyVBMC itself continues to support Python 3.10 and newer; on Python 3.10 or 3.11, pip does not install the PyMC dependencies because of their Python version marker, and PyMCTarget is unavailable. With Conda, install the named packages directly:

conda install --channel=conda-forge "pyvbmc>=1.5" pymc

The bracketed torch, arviz and pymc names are pip dependency groups; they are not Conda package names. All three integrations are optional and loaded on demand.

Examples and tests

Install Jupyter Notebook to run the examples. You can skip this step if your environment already has Jupyter Notebook, but an unrelated jupyter executable on your path can cause import errors:

python -m pip install notebook

or, with Conda:

conda install --channel=conda-forge jupyter

Open the example notebooks with:

python -m pyvbmc

Example 2 uses plotly for one interactive figure. Install it with:

python -m pip install --upgrade "pyvbmc[examples]"

PyVBMC Example 9: Torch and JAX models and posterior exports uses both posterior exports and requires Python 3.12 or newer. A CPU-only environment is sufficient. Install its PyVBMC dependencies with:

python -m pip install --upgrade "pyvbmc[torch,arviz]"

Install JAX separately to run the short JAX target section, following the JAX installation guide. JAX is not required for the Torch target or posterior-export sections.

The test suite is included in the source distribution and repository. To run it, follow the Installation instructions for developers, then run this command from the PyVBMC repository root:

python -m pytest --reruns=5 -x

The --reruns=5 argument retries a failed test up to five times because some tests are stochastic. The complete suite can take 20–30 minutes or longer, depending on the hardware.

To install directly from the latest source, see the Installation instructions for developers.