BayesFlux

BayesFlux is a JAX Python package that provides Fast Linear algebra sUbspace eXtraction for Bayesian inverse problems built off https://github.com/joshuawchen/randLAX.

Features

• Active Subspace for parameter dimension reduction
• Informative Output Subspace for data dimension reduction

Installation

Core Installation (JAX-only functionality)

You can install the core BayesFlux package using pip:

pip install bayesflux

This installs the JAX-based functionality only and does not require Fenics or hippylib.

Installation with hippylib Support (Requires Fenics)

Some BayesFlux functionality depends on hippylib, which requires Fenics 2019.1.

Fenics has system-level dependencies and cannot be reliably installed via pip alone. You must first create a conda environment.

Step 1 — Create a Fenics environment

conda create -n fenics-2019.1_env -c conda-forge fenics==2019.1.0
conda activate fenics-2019.1_env

Step 2 — Install BayesFlux with the hippylib extra

pip install bayesflux[hippylib]

This installs:

• hippylib
• hickle
• bayesflux

Make sure the conda environment is activated before running pip install.

For Developers

If your software depends on BayesFlux with hippylib functionality, declare the dependency in your pyproject.toml as:

dependencies = [
    "bayesflux[hippylib]>=<minimum_version>"
]

However, your users must still create the Fenics conda environment before installing your software:

conda create -n fenics-2019.1_env -c conda-forge fenics==2019.1.0
conda activate fenics-2019.1_env
pip install your_package

Important: pip cannot install Fenics. Any software depending on bayesflux[hippylib] must document the required conda environment setup.

Requirements

• Python 3.9+