BayesCaliFramework

Python research code for Bayesian calibration of computer simulators with Gaussian-process emulators and Gaussian-process discrepancy models.

The framework contains modular implementations of standard KOH calibration, scaled-discrepancy calibration, orthogonal discrepancy calibration, randomized tangent-space calibration, probabilistic orthogonal Gaussian closure, and posterior-driven active-learning utilities.

Repository Layout

BayesCaliFramework/
  ActiveLearning/   Acquisition functions for posterior-driven simulator design
  Discrepancy/      GP discrepancy kernels and orthogonal covariance operators
  Inference/        Likelihoods, priors, calibrators, and TMCMC samplers
  Simulator/        Scalar and field-valued GP emulator utilities
  Wrapper/          Higher-level calibration interfaces
  demo/             Reproducible example scripts
  illustration/     Small scripts for conceptual figures

Generated outputs are intentionally not tracked. Most demo scripts write their results under a local outputs/ directory.

Installation

Clone the repository and install it in editable mode:

git clone https://github.com/bominwang/BayesCaliFramework-Code.git
cd BayesCaliFramework-Code
python -m pip install -e .

The core dependencies are listed in requirements.txt. Installing PyTorch may require choosing the wheel that matches your local CPU/GPU environment.

Minimal Usage

The high-level wrappers expose the main calibration methods:

from BayesCaliFramework.Wrapper import KOH, OKOH, SKOH, AngularRTKOH
from BayesCaliFramework.Simulator.gp_model import GPModelConfig
from BayesCaliFramework.Simulator.gp_training import ScalarEmulator, ScalarEmulatorConfig

Typical workflow:

1. Generate or load simulator runs.
2. Train a GP emulator with ScalarEmulator or FieldEmulator.
3. Choose a calibration wrapper such as KOH, OKOH, SKOH, or AngularRTKOH.
4. Run posterior inference with TMCMC.
5. Use the posterior samples for prediction, diagnostics, or active learning.

Demo Scripts

Example scripts are available under BayesCaliFramework/demo/. For example:

python BayesCaliFramework/demo/case01/run_koh.py --help
python BayesCaliFramework/demo/case01/run_okoh.py --help
python BayesCaliFramework/demo/case01/run_rtkoh_pogc.py --help

The demos are designed as executable research examples. They may take a long time on CPU-only machines because GP training and TMCMC sampling are both computationally intensive.

Notes

• This repository contains source code and runnable example scripts only.
• Simulation outputs, posterior samples, generated figures, manuscript files, verification scratch files, and reference PDFs are excluded.
• The code is research-oriented and APIs may change as methods are refined.