Welcome to MUQ (pronounced “muck”), a modular software framework for defining and solving forward and inverse uncertainty quantification problems.

Purpose

Uncertainty quantification (UQ) is important in many different applications. MUQ aims to make advanced probabilistic UQ tools easy to use in either c++ or python, and enable cutting-edge method development through its modular structure.

MUQ has a variety of capabilities, including:

Various Markov chain Monte Carlo methods
Graphical modeling with a mix of statistical and physical components.
Gaussian processes
Karhunen Loève expansions.
Transport maps
Nonlinear Optimization
Generalized Polynomial Chaos Expansions

Installation:

MUQ is available on Linux and OSX as a conda package, docker image, or from source. For many users, getting started can be as easy as running

conda install -c conda-forge muq

For more installation options, check out the installation guide.

Getting Started

MUQ is composed of several different modules, which work together to define and solve UQ problems. Documentation for each of these modules is included with our doxygen-generated API documentation. Most applications will require using the modeling module to define statistical models or interact with user-defined models. Learning the basics of this module is therefore a good place to start.

Interested in forward UQ?

First, get acquainted with the modeling module. You'll need to use one or more instances of the ModPiece class to define the model that will be evaluated by the UQ algorithm.
Once you have a model, check out the polynomial chaos module.
Other examples can be found by selecting the "PCE" examples on the MUQ webpage.

Want to tackle Bayesian inverse problems?

Just like for forward UQ, you'll want to get familiar with the modeling module module to define a forward model. The WorkGraph class within the modeling module is also used to combine multiple components (e.g., the prior, forward model, and likelihood function) comprising the Bayesian posterior distribution.
Look at methods in the sampling algorithms module to generate samples of your Bayesian posterior.
Other examples can be found by filtering the "MCMC" examples on the MUQ webpage.

You can also find many examples using both the c++ and Python interfaces to MUQ. These examples can provide useful starting places for using MUQ on your own problems.

Getting Connected

Join the MUQ Slack channel via our website to get in touch with MUQ developers and other users. We are always happy to help!

Citing

When publishing work based on MUQ, please cite our publication in the Journal of Open Source Software.

@article{Parno2021,
  doi = {10.21105/joss.03076},
  url = {https://doi.org/10.21105/joss.03076},
  year = {2021},
  publisher = {The Open Journal},
  volume = {6},
  number = {68},
  pages = {3076},
  author = {Matthew Parno and Andrew Davis and Linus Seelinger},
  title = {MUQ: The MIT Uncertainty Quantification Library},
  journal = {Journal of Open Source Software}
}

Contributing

Want to help develop MUQ?

Yes, please! We frequently discuss future developments on Slack (join via our website), so feel free to drop by! Then fork the muq2 repository and submit a pull request when ready. Also check out our style guide.