Skip to content

Quickstart

This page is command-first. If you want a working run with minimal setup, follow sections 1-3 in order.

1. Install

Prerequisites:

  • Python 3.10
  • C/C++ toolchain (tinyDA/GPy build dependencies)
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 ".[dev]"

macOS note:

xcode-select --install
brew install python@3.10

If your shell does not find pytest, run tests as:

python -m pytest -q

2. Run the toy forward workflow

python examples/toy_problem/run_forward_toy.py

This script:

  • generates synthetic snapshots from the toy HF model,
  • fits a POD basis and a multi-output GP,
  • prints validation metrics (RMSE and coverage),
  • produces diagnostic plots.

3. Run the toy inverse workflow

python examples/toy_problem/run_backward_toy.py

This script runs both:

  • single-posterior active MCMC, and
  • delayed-acceptance active MCMC with adaptive subchain updates.

It also plots posterior samples, HF usage, and subchain behavior.

4. Change run settings

Both toy scripts are notebook-style Python files with configuration constants near the top. Edit those constants directly:

  • examples/toy_problem/run_forward_toy.py for snapshot count, POD rank, GP kernel
  • examples/toy_problem/run_backward_toy.py for chain length, burn-in, thresholds, and adaptive controls

There is currently no CLI argument parser for these scripts.

5. Use the docs notebooks

For step-by-step narrative versions of the same workflows:

  • docs/tutorials/forward_toy_notebook.py
  • docs/tutorials/backward_toy_notebook.py

Serve docs locally:

python -m pip install -e ".[docs]"
mkdocs serve

6. Next steps

  • Read Concepts for the model and algorithm details.
  • Open the API reference pages for module-level docs.
  • For the PDE benchmark, see examples/navier_stokes/ and the Navier-Stokes tutorial page.