Welcome to SEPIA!¶
What is SEPIA?¶
SEPIA (Simulation-Enabled Prediction, Inference, and Analysis) implements Bayesian emulation and calibration with the ability to handle multivariate outputs. It is based on the Matlab code GPMSA. For more, see About SEPIA.
Example code is described in Examples. Full source code is available on GitHub. While SEPIA is still under development, users should pull the newest code from Github frequently. If you have installed using the instructions below, you should not need to reinstall after pulling new code.
SEPIA is still under development and is intended to be a research tool, not production-level code. Please submit questions as GitHub issues if you encounter unexpected behavior or need help using SEPIA. As of the current release, basic functionality (model setup, likelihood evaluation, MCMC sampling, multivariate calibration, visualization and diagnostics, predictions) is complete and has been tested.
Installation¶
For cleaner package management and to avoid conflicts between different versions of packages, we recommend installing inside an Anaconda or pip environment (see Conda docs for details on Anaconda environments). However, this is not required.
First, pull down the current source code from GitHub either by downloading a zip file or using git clone. If you prefer, you can download the latest stable release instead of the master branch.
From the command line, while in the main SEPIA directory, use the following command to install sepia:
pip install -e .[sepia]
The -e flag signals developer mode, meaning that if you update the code from Github, your installation will automatically take those changes into account without requiring re-installation. Some other essential packages used in SEPIA may be installed if they do not exist in your system or environment.
If you encounter problems with the above install method, you may try to install dependencies manually before installing SEPIA. First, ensure you have a recent version of Python (greater than 3.5). Then, install packages numpy, scipy, pandas, matplotlib, seaborn, statsmodels, and tqdm.
Resources for new users¶
Before getting started, we highly recommend reading the General workflow guide. This goes through the general workflow for setting up a model, doing MCMC, and checking results.
If you are familiar with the general workflow but need a quick reference for common tasks, see Helpful Code Snippets.
If you are a GPMSA user, we suggest reading SEPIA for GPMSA users.
Citing Sepia¶
Using Sepia in your work? Please cite it as:
James Gattiker, Natalie Klein, Earl Lawrence, & Grant Hutchings. lanl/SEPIA. Zenodo. https://doi.org/10.5281/zenodo.3979584