gp_active_mcmc

gp_active_mcmc couples Gaussian-process surrogates with active/adaptive MCMC to accelerate Bayesian inference when high-fidelity forward models are expensive. The library lets you:

• build POD–GP surrogates from snapshot data,
• run delayed-acceptance or adaptive-subchain samplers backed by tinyDA,
• instrument HF usage, surrogate error, and posterior summaries for publications.

---

Pick your starting point

Workflow	Where to look	Runtime budget
Surrogate-only experimentation	Tutorials → Forward toy	seconds
Active / adaptive inference demos	Tutorials → Backward toy	minutes
CFD benchmark (FEniCS/DOLFINx)	Tutorials → Navier-Stokes & examples/navier_stokes/	hours

---

Install quickstart

The tested stack (pinned for tinyDA/GPy stability):

• Python 3.10.19, NumPy 1.26.4, SciPy 1.12.0
• scikit-learn 1.7.2, tinyDA 0.9.21
• Ray 2.53.0 (indirect requirement of tinyDA)
• GPy 1.13.2

git clone https://github.com/filippozacchei/MFDA_ActiveLearning.git
cd MFDA_ActiveLearning
python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e .
python examples/toy_problem/run_forward_toy.py

macOS-specific setup (Apple Silicon and Intel):

xcode-select --install
brew install python@3.10
git clone https://github.com/filippozacchei/MFDA_ActiveLearning.git
cd MFDA_ActiveLearning
python3.10 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip setuptools wheel
python -m pip install -e .

Documentation stack:

python -m pip install -e ".[docs]"
mkdocs serve  # open http://127.0.0.1:8000

Testing stack (optional):

python -m pip install -e ".[test]"
python -m pytest -q

Navier–Stokes scripts additionally require a conda-forge environment with fenics-dolfinx=0.9.0, mpich, gmsh=4.15.0, python-gmsh=4.15.0, and pyvista.

---

Documentation map

• User Guide → conceptual overview and workflow rationale.
• Tutorials → executable notebooks for toy and CFD examples (rendered via mkdocs-jupyter).
• API Reference → autogenerated docs for gp_active_mcmc.surrogates, .inference, .diagnostics, and .utils.

---

Community & governance

• Contributions follow CONTRIBUTING and are linted/tested via nox.
• Behavior in GitHub issues/PRs is governed by the Code of Conduct.
• Citation details live in CITATION.cff.

References

• MCMC-guided active learning with Gaussian-process surrogates — primary reference for the active-learning scheme adopted here.
• Delayed-acceptance MLDA / tinyDA framework — details the underlying sampler infrastructure we rely on.

Ready to dive in? Start with the Quickstart guide below.

Next: User Guide → Quickstart