This repository contains materials for an introductory course on "Surrogate Modeling and Design Optimization," originally developed for the Shared Education in Artificial intelligence and Machine learning (SEAM) professional development program at Lawrence Livermore National Laboratory (LLNL).
It features the surmod Python package, which implements core techniques including Feedforward Neural Networks (FFNNs) and Gaussian Processes (GPs) as surrogates, basic Bayesian Optimization (BO) for design optimization, and Sensitivity Analysis (SA).
Weekly workflow scripts in the scripts directory demonstrate these methods in practice.
Example datasets included in the data directory are used to support hands-on learning and experimentation.
For more detailed documentation on the modules in the surmod package, please view our documentation.
Information about the SEAM program is available to employees of Lawrence Livermore National Laboratory at https://seam.llnl.gov.
Please send correspondence about this code to seam@llnl.gov.
- Clone the repository
git clone https://github.com/LLNL/SEAMsurrogates.git
cd SEAMsurrogates- Create a virtual environment
python3 -m venv .venv- Activate the virtual environment
source .venv/bin/activate- Install the
surmodpackage and dependencies locally
pip install -e .- For development (optional): Install development dependencies for testing and linting
pip install -e ".[dev]"The datasets in the data directory are from different sources:
- The JAG ICF dataset is from this repo.
- The borehole dataset was simulated from the borehole function.
- The datasets that start with
hstare from simulations of the Hubble Space Telescope (HST), and are sampled from larger datasets available here.
| Week | Topic | Driver Scripts |
|---|---|---|
| 1 | Deep Learning, PyTorch, and Surrogate Modeling | |
| 2 | Neural Network (NN) Surrogate Models | nn_sandbox.py |
| 3 | NN Surrogates with Example Data | nn_fromdata.py |
| 4 | Gaussian Process (GP) Surrogate Models | gp_sandbox.py |
| 5 | GP Surrogates with Example Data | gp_fromdata.py |
| 6 | Bayesian Optimization (BO) for Design | bo_sandbox.py |
| 7 | BO with Example Data | bo_fromdata.py |
| 8 | Sensitivity Analysis (SA) | sa_sandbox.py, sa_fromdata.py |
| 9 | Final Project Problem Formulation | |
| 10 | Final Project Development | |
| 11 | Final Project Development | |
| 12 | Final Project Presentations |
- Jason Bernstein (bernstein8@llnl.gov)
- Emily Bogle (bogle4@llnl.gov)
- Andrew Gillette (gillette7@llnl.gov)
- Kevin Quinlan (quinlan5@llnl.gov)
This software is distributed under the terms of the BSD-Commercial license.
See LICENSE and NOTICE for details.
LLNL-CODE-2010455